code stringlengths 82 54.1k | code_codestyle int64 0 699 | style_context stringlengths 111 35.6k | style_context_codestyle int64 0 699 | label int64 0 1 |
|---|---|---|---|---|
import copy
from collections import OrderedDict
from typing import Dict, Mapping
from packaging import version
from ...configuration_utils import PretrainedConfig
from ...onnx import OnnxConfig
from ...utils import logging
from ..auto import CONFIG_MAPPING
__a :Optional[int] = logging.get_logger(__name__)
__a :Optional[int] = {
'facebook/detr-resnet-50': 'https://huggingface.co/facebook/detr-resnet-50/resolve/main/config.json',
# See all DETR models at https://huggingface.co/models?filter=detr
}
class _a ( snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Dict = 'detr'
_lowerCamelCase : Union[str, Any] = ['past_key_values']
_lowerCamelCase : Optional[int] = {
'hidden_size': 'd_model',
'num_attention_heads': 'encoder_attention_heads',
}
def __init__( self : int , UpperCAmelCase : List[Any]=True , UpperCAmelCase : Union[str, Any]=None , UpperCAmelCase : Dict=3 , UpperCAmelCase : List[Any]=100 , UpperCAmelCase : Union[str, Any]=6 , UpperCAmelCase : Optional[int]=2048 , UpperCAmelCase : str=8 , UpperCAmelCase : Optional[Any]=6 , UpperCAmelCase : List[Any]=2048 , UpperCAmelCase : Tuple=8 , UpperCAmelCase : List[Any]=0.0 , UpperCAmelCase : Any=0.0 , UpperCAmelCase : Tuple=True , UpperCAmelCase : List[str]="relu" , UpperCAmelCase : List[str]=256 , UpperCAmelCase : str=0.1 , UpperCAmelCase : Optional[Any]=0.0 , UpperCAmelCase : Any=0.0 , UpperCAmelCase : Optional[int]=0.02 , UpperCAmelCase : Tuple=1.0 , UpperCAmelCase : Any=False , UpperCAmelCase : Union[str, Any]="sine" , UpperCAmelCase : str="resnet50" , UpperCAmelCase : List[Any]=True , UpperCAmelCase : int=False , UpperCAmelCase : Dict=1 , UpperCAmelCase : Tuple=5 , UpperCAmelCase : Optional[int]=2 , UpperCAmelCase : Dict=1 , UpperCAmelCase : Union[str, Any]=1 , UpperCAmelCase : Tuple=5 , UpperCAmelCase : Optional[Any]=2 , UpperCAmelCase : Dict=0.1 , **UpperCAmelCase : int , ):
if backbone_config is not None and use_timm_backbone:
raise ValueError("You can't specify both `backbone_config` and `use_timm_backbone`." )
if not use_timm_backbone:
if backbone_config is None:
logger.info("`backbone_config` is `None`. Initializing the config with the default `ResNet` backbone." )
A_ = CONFIG_MAPPING["resnet"](out_features=["stage4"] )
elif isinstance(UpperCAmelCase , UpperCAmelCase ):
A_ = backbone_config.get("model_type" )
A_ = CONFIG_MAPPING[backbone_model_type]
A_ = config_class.from_dict(UpperCAmelCase )
# set timm attributes to None
A_ , A_ , A_ = None, None, None
A_ = use_timm_backbone
A_ = backbone_config
A_ = num_channels
A_ = num_queries
A_ = d_model
A_ = encoder_ffn_dim
A_ = encoder_layers
A_ = encoder_attention_heads
A_ = decoder_ffn_dim
A_ = decoder_layers
A_ = decoder_attention_heads
A_ = dropout
A_ = attention_dropout
A_ = activation_dropout
A_ = activation_function
A_ = init_std
A_ = init_xavier_std
A_ = encoder_layerdrop
A_ = decoder_layerdrop
A_ = encoder_layers
A_ = auxiliary_loss
A_ = position_embedding_type
A_ = backbone
A_ = use_pretrained_backbone
A_ = dilation
# Hungarian matcher
A_ = class_cost
A_ = bbox_cost
A_ = giou_cost
# Loss coefficients
A_ = mask_loss_coefficient
A_ = dice_loss_coefficient
A_ = bbox_loss_coefficient
A_ = giou_loss_coefficient
A_ = eos_coefficient
super().__init__(is_encoder_decoder=UpperCAmelCase , **UpperCAmelCase )
@property
def __A ( self : Dict ):
return self.encoder_attention_heads
@property
def __A ( self : List[Any] ):
return self.d_model
@classmethod
def __A ( cls : Dict , UpperCAmelCase : PretrainedConfig , **UpperCAmelCase : Any ):
return cls(backbone_config=UpperCAmelCase , **UpperCAmelCase )
def __A ( self : int ):
A_ = copy.deepcopy(self.__dict__ )
if output["backbone_config"] is not None:
A_ = self.backbone_config.to_dict()
A_ = self.__class__.model_type
return output
class _a ( snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Tuple = version.parse('1.11' )
@property
def __A ( self : List[str] ):
return OrderedDict(
[
("pixel_values", {0: "batch", 1: "num_channels", 2: "height", 3: "width"}),
("pixel_mask", {0: "batch"}),
] )
@property
def __A ( self : Tuple ):
return 1E-5
@property
def __A ( self : Union[str, Any] ):
return 12 | 86 |
import argparse
import json
from pathlib import Path
import requests
import timm
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from transformers import DeiTImageProcessor, ViTConfig, ViTForImageClassification, ViTImageProcessor, ViTModel
from transformers.utils import logging
logging.set_verbosity_info()
__a :Dict = logging.get_logger(__name__)
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : Tuple=False ):
"""simple docstring"""
A_ = []
for i in range(config.num_hidden_layers ):
# encoder layers: output projection, 2 feedforward neural networks and 2 layernorms
rename_keys.append((f'''blocks.{i}.norm1.weight''', f'''vit.encoder.layer.{i}.layernorm_before.weight''') )
rename_keys.append((f'''blocks.{i}.norm1.bias''', f'''vit.encoder.layer.{i}.layernorm_before.bias''') )
rename_keys.append((f'''blocks.{i}.attn.proj.weight''', f'''vit.encoder.layer.{i}.attention.output.dense.weight''') )
rename_keys.append((f'''blocks.{i}.attn.proj.bias''', f'''vit.encoder.layer.{i}.attention.output.dense.bias''') )
rename_keys.append((f'''blocks.{i}.norm2.weight''', f'''vit.encoder.layer.{i}.layernorm_after.weight''') )
rename_keys.append((f'''blocks.{i}.norm2.bias''', f'''vit.encoder.layer.{i}.layernorm_after.bias''') )
rename_keys.append((f'''blocks.{i}.mlp.fc1.weight''', f'''vit.encoder.layer.{i}.intermediate.dense.weight''') )
rename_keys.append((f'''blocks.{i}.mlp.fc1.bias''', f'''vit.encoder.layer.{i}.intermediate.dense.bias''') )
rename_keys.append((f'''blocks.{i}.mlp.fc2.weight''', f'''vit.encoder.layer.{i}.output.dense.weight''') )
rename_keys.append((f'''blocks.{i}.mlp.fc2.bias''', f'''vit.encoder.layer.{i}.output.dense.bias''') )
# projection layer + position embeddings
rename_keys.extend(
[
("cls_token", "vit.embeddings.cls_token"),
("patch_embed.proj.weight", "vit.embeddings.patch_embeddings.projection.weight"),
("patch_embed.proj.bias", "vit.embeddings.patch_embeddings.projection.bias"),
("pos_embed", "vit.embeddings.position_embeddings"),
] )
if base_model:
# layernorm + pooler
rename_keys.extend(
[
("norm.weight", "layernorm.weight"),
("norm.bias", "layernorm.bias"),
("pre_logits.fc.weight", "pooler.dense.weight"),
("pre_logits.fc.bias", "pooler.dense.bias"),
] )
# if just the base model, we should remove "vit" from all keys that start with "vit"
A_ = [(pair[0], pair[1][4:]) if pair[1].startswith("vit" ) else pair for pair in rename_keys]
else:
# layernorm + classification head
rename_keys.extend(
[
("norm.weight", "vit.layernorm.weight"),
("norm.bias", "vit.layernorm.bias"),
("head.weight", "classifier.weight"),
("head.bias", "classifier.bias"),
] )
return rename_keys
def __snake_case ( __UpperCamelCase : Optional[int] ,__UpperCamelCase : Tuple ,__UpperCamelCase : Any=False ):
"""simple docstring"""
for i in range(config.num_hidden_layers ):
if base_model:
A_ = ""
else:
A_ = "vit."
# read in weights + bias of input projection layer (in timm, this is a single matrix + bias)
A_ = state_dict.pop(f'''blocks.{i}.attn.qkv.weight''' )
A_ = state_dict.pop(f'''blocks.{i}.attn.qkv.bias''' )
# next, add query, keys and values (in that order) to the state dict
A_ = in_proj_weight[
: config.hidden_size, :
]
A_ = in_proj_bias[: config.hidden_size]
A_ = in_proj_weight[
config.hidden_size : config.hidden_size * 2, :
]
A_ = in_proj_bias[
config.hidden_size : config.hidden_size * 2
]
A_ = in_proj_weight[
-config.hidden_size :, :
]
A_ = in_proj_bias[-config.hidden_size :]
def __snake_case ( __UpperCamelCase : List[Any] ):
"""simple docstring"""
A_ = ["head.weight", "head.bias"]
for k in ignore_keys:
state_dict.pop(__UpperCamelCase ,__UpperCamelCase )
def __snake_case ( __UpperCamelCase : Any ,__UpperCamelCase : Optional[int] ,__UpperCamelCase : List[str] ):
"""simple docstring"""
A_ = dct.pop(__UpperCamelCase )
A_ = val
def __snake_case ( ):
"""simple docstring"""
A_ = "http://images.cocodataset.org/val2017/000000039769.jpg"
A_ = Image.open(requests.get(__UpperCamelCase ,stream=__UpperCamelCase ).raw )
return im
@torch.no_grad()
def __snake_case ( __UpperCamelCase : Union[str, Any] ,__UpperCamelCase : Optional[int] ):
"""simple docstring"""
A_ = ViTConfig()
A_ = False
# dataset (ImageNet-21k only or also fine-tuned on ImageNet 2012), patch_size and image_size
if vit_name[-5:] == "in21k":
A_ = True
A_ = int(vit_name[-12:-10] )
A_ = int(vit_name[-9:-6] )
else:
A_ = 1000
A_ = "huggingface/label-files"
A_ = "imagenet-1k-id2label.json"
A_ = json.load(open(hf_hub_download(__UpperCamelCase ,__UpperCamelCase ,repo_type="dataset" ) ,"r" ) )
A_ = {int(__UpperCamelCase ): v for k, v in idalabel.items()}
A_ = idalabel
A_ = {v: k for k, v in idalabel.items()}
A_ = int(vit_name[-6:-4] )
A_ = int(vit_name[-3:] )
# size of the architecture
if "deit" in vit_name:
if vit_name[9:].startswith("tiny" ):
A_ = 192
A_ = 768
A_ = 12
A_ = 3
elif vit_name[9:].startswith("small" ):
A_ = 384
A_ = 1536
A_ = 12
A_ = 6
else:
pass
else:
if vit_name[4:].startswith("small" ):
A_ = 768
A_ = 2304
A_ = 8
A_ = 8
elif vit_name[4:].startswith("base" ):
pass
elif vit_name[4:].startswith("large" ):
A_ = 1024
A_ = 4096
A_ = 24
A_ = 16
elif vit_name[4:].startswith("huge" ):
A_ = 1280
A_ = 5120
A_ = 32
A_ = 16
# load original model from timm
A_ = timm.create_model(__UpperCamelCase ,pretrained=__UpperCamelCase )
timm_model.eval()
# load state_dict of original model, remove and rename some keys
A_ = timm_model.state_dict()
if base_model:
remove_classification_head_(__UpperCamelCase )
A_ = create_rename_keys(__UpperCamelCase ,__UpperCamelCase )
for src, dest in rename_keys:
rename_key(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
read_in_q_k_v(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
# load HuggingFace model
if vit_name[-5:] == "in21k":
A_ = ViTModel(__UpperCamelCase ).eval()
else:
A_ = ViTForImageClassification(__UpperCamelCase ).eval()
model.load_state_dict(__UpperCamelCase )
# Check outputs on an image, prepared by ViTImageProcessor/DeiTImageProcessor
if "deit" in vit_name:
A_ = DeiTImageProcessor(size=config.image_size )
else:
A_ = ViTImageProcessor(size=config.image_size )
A_ = image_processor(images=prepare_img() ,return_tensors="pt" )
A_ = encoding["pixel_values"]
A_ = model(__UpperCamelCase )
if base_model:
A_ = timm_model.forward_features(__UpperCamelCase )
assert timm_pooled_output.shape == outputs.pooler_output.shape
assert torch.allclose(__UpperCamelCase ,outputs.pooler_output ,atol=1E-3 )
else:
A_ = timm_model(__UpperCamelCase )
assert timm_logits.shape == outputs.logits.shape
assert torch.allclose(__UpperCamelCase ,outputs.logits ,atol=1E-3 )
Path(__UpperCamelCase ).mkdir(exist_ok=__UpperCamelCase )
print(f'''Saving model {vit_name} to {pytorch_dump_folder_path}''' )
model.save_pretrained(__UpperCamelCase )
print(f'''Saving image processor to {pytorch_dump_folder_path}''' )
image_processor.save_pretrained(__UpperCamelCase )
if __name__ == "__main__":
__a :str = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
'--vit_name',
default='vit_base_patch16_224',
type=str,
help='Name of the ViT timm model you\'d like to convert.',
)
parser.add_argument(
'--pytorch_dump_folder_path', default=None, type=str, help='Path to the output PyTorch model directory.'
)
__a :Optional[int] = parser.parse_args()
convert_vit_checkpoint(args.vit_name, args.pytorch_dump_folder_path) | 86 | 1 |
import unittest
import numpy as np
import torch
from diffusers import DDIMPipeline, DDIMScheduler, UNetaDModel
from diffusers.utils.testing_utils import enable_full_determinism, require_torch_gpu, slow, torch_device
from ..pipeline_params import UNCONDITIONAL_IMAGE_GENERATION_BATCH_PARAMS, UNCONDITIONAL_IMAGE_GENERATION_PARAMS
from ..test_pipelines_common import PipelineTesterMixin
enable_full_determinism()
class _a ( snake_case_ , unittest.TestCase ):
"""simple docstring"""
_lowerCamelCase : int = DDIMPipeline
_lowerCamelCase : Union[str, Any] = UNCONDITIONAL_IMAGE_GENERATION_PARAMS
_lowerCamelCase : Any = PipelineTesterMixin.required_optional_params - {
'num_images_per_prompt',
'latents',
'callback',
'callback_steps',
}
_lowerCamelCase : List[Any] = UNCONDITIONAL_IMAGE_GENERATION_BATCH_PARAMS
_lowerCamelCase : Optional[int] = False
def __A ( self : Any ):
torch.manual_seed(0 )
A_ = UNetaDModel(
block_out_channels=(32, 64) , layers_per_block=2 , sample_size=32 , in_channels=3 , out_channels=3 , down_block_types=("DownBlock2D", "AttnDownBlock2D") , up_block_types=("AttnUpBlock2D", "UpBlock2D") , )
A_ = DDIMScheduler()
A_ = {"unet": unet, "scheduler": scheduler}
return components
def __A ( self : Optional[Any] , UpperCAmelCase : Optional[int] , UpperCAmelCase : Optional[int]=0 ):
if str(UpperCAmelCase ).startswith("mps" ):
A_ = torch.manual_seed(UpperCAmelCase )
else:
A_ = torch.Generator(device=UpperCAmelCase ).manual_seed(UpperCAmelCase )
A_ = {
"batch_size": 1,
"generator": generator,
"num_inference_steps": 2,
"output_type": "numpy",
}
return inputs
def __A ( self : str ):
A_ = "cpu"
A_ = self.get_dummy_components()
A_ = self.pipeline_class(**UpperCAmelCase )
pipe.to(UpperCAmelCase )
pipe.set_progress_bar_config(disable=UpperCAmelCase )
A_ = self.get_dummy_inputs(UpperCAmelCase )
A_ = pipe(**UpperCAmelCase ).images
A_ = image[0, -3:, -3:, -1]
self.assertEqual(image.shape , (1, 32, 32, 3) )
A_ = np.array(
[1.000E00, 5.717E-01, 4.717E-01, 1.000E00, 0.000E00, 1.000E00, 3.000E-04, 0.000E00, 9.000E-04] )
A_ = np.abs(image_slice.flatten() - expected_slice ).max()
self.assertLessEqual(UpperCAmelCase , 1E-3 )
def __A ( self : Any ):
super().test_dict_tuple_outputs_equivalent(expected_max_difference=3E-3 )
def __A ( self : Optional[int] ):
super().test_save_load_local(expected_max_difference=3E-3 )
def __A ( self : Optional[int] ):
super().test_save_load_optional_components(expected_max_difference=3E-3 )
def __A ( self : Any ):
super().test_inference_batch_single_identical(expected_max_diff=3E-3 )
@slow
@require_torch_gpu
class _a ( unittest.TestCase ):
"""simple docstring"""
def __A ( self : Any ):
A_ = "google/ddpm-cifar10-32"
A_ = UNetaDModel.from_pretrained(UpperCAmelCase )
A_ = DDIMScheduler()
A_ = DDIMPipeline(unet=UpperCAmelCase , scheduler=UpperCAmelCase )
ddim.to(UpperCAmelCase )
ddim.set_progress_bar_config(disable=UpperCAmelCase )
A_ = torch.manual_seed(0 )
A_ = ddim(generator=UpperCAmelCase , eta=0.0 , output_type="numpy" ).images
A_ = image[0, -3:, -3:, -1]
assert image.shape == (1, 32, 32, 3)
A_ = np.array([0.1_723, 0.1_617, 0.1_600, 0.1_626, 0.1_497, 0.1_513, 0.1_505, 0.1_442, 0.1_453] )
assert np.abs(image_slice.flatten() - expected_slice ).max() < 1E-2
def __A ( self : Optional[Any] ):
A_ = "google/ddpm-ema-bedroom-256"
A_ = UNetaDModel.from_pretrained(UpperCAmelCase )
A_ = DDIMScheduler.from_pretrained(UpperCAmelCase )
A_ = DDIMPipeline(unet=UpperCAmelCase , scheduler=UpperCAmelCase )
ddpm.to(UpperCAmelCase )
ddpm.set_progress_bar_config(disable=UpperCAmelCase )
A_ = torch.manual_seed(0 )
A_ = ddpm(generator=UpperCAmelCase , output_type="numpy" ).images
A_ = image[0, -3:, -3:, -1]
assert image.shape == (1, 256, 256, 3)
A_ = np.array([0.0_060, 0.0_201, 0.0_344, 0.0_024, 0.0_018, 0.0_002, 0.0_022, 0.0_000, 0.0_069] )
assert np.abs(image_slice.flatten() - expected_slice ).max() < 1E-2 | 86 |
def __snake_case ( __UpperCamelCase : int = 50 ):
"""simple docstring"""
A_ = [1] * (length + 1)
for row_length in range(length + 1 ):
for tile_length in range(2 ,5 ):
for tile_start in range(row_length - tile_length + 1 ):
ways_number[row_length] += ways_number[
row_length - tile_start - tile_length
]
return ways_number[length]
if __name__ == "__main__":
print(F"{solution() = }") | 86 | 1 |
from PIL import Image
def __snake_case ( __UpperCamelCase : Image ,__UpperCamelCase : int ):
"""simple docstring"""
A_ = (259 * (level + 255)) / (255 * (259 - level))
def contrast(__UpperCamelCase : int ) -> int:
return int(128 + factor * (c - 128) )
return img.point(__UpperCamelCase )
if __name__ == "__main__":
# Load image
with Image.open('image_data/lena.jpg') as img:
# Change contrast to 170
__a :Tuple = change_contrast(img, 170)
cont_img.save('image_data/lena_high_contrast.png', format='png') | 86 |
from typing import Any, Dict, List, Union
from ..utils import add_end_docstrings, is_torch_available, is_vision_available, logging, requires_backends
from .base import PIPELINE_INIT_ARGS, ChunkPipeline
if is_vision_available():
from PIL import Image
from ..image_utils import load_image
if is_torch_available():
import torch
from transformers.modeling_outputs import BaseModelOutput
from ..models.auto.modeling_auto import MODEL_FOR_ZERO_SHOT_OBJECT_DETECTION_MAPPING
__a :List[str] = logging.get_logger(__name__)
@add_end_docstrings(snake_case_ )
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Any , **UpperCAmelCase : List[str] ):
super().__init__(**UpperCAmelCase )
if self.framework == "tf":
raise ValueError(f'''The {self.__class__} is only available in PyTorch.''' )
requires_backends(self , "vision" )
self.check_model_type(UpperCAmelCase )
def __call__( self : Optional[int] , UpperCAmelCase : Union[str, "Image.Image", List[Dict[str, Any]]] , UpperCAmelCase : Union[str, List[str]] = None , **UpperCAmelCase : List[Any] , ):
if "text_queries" in kwargs:
A_ = kwargs.pop("text_queries" )
if isinstance(UpperCAmelCase , (str, Image.Image) ):
A_ = {"image": image, "candidate_labels": candidate_labels}
else:
A_ = image
A_ = super().__call__(UpperCAmelCase , **UpperCAmelCase )
return results
def __A ( self : int , **UpperCAmelCase : Tuple ):
A_ = {}
if "threshold" in kwargs:
A_ = kwargs["threshold"]
if "top_k" in kwargs:
A_ = kwargs["top_k"]
return {}, {}, postprocess_params
def __A ( self : List[str] , UpperCAmelCase : Dict ):
A_ = load_image(inputs["image"] )
A_ = inputs["candidate_labels"]
if isinstance(UpperCAmelCase , UpperCAmelCase ):
A_ = candidate_labels.split("," )
A_ = torch.tensor([[image.height, image.width]] , dtype=torch.intaa )
for i, candidate_label in enumerate(UpperCAmelCase ):
A_ = self.tokenizer(UpperCAmelCase , return_tensors=self.framework )
A_ = self.image_processor(UpperCAmelCase , return_tensors=self.framework )
yield {
"is_last": i == len(UpperCAmelCase ) - 1,
"target_size": target_size,
"candidate_label": candidate_label,
**text_inputs,
**image_features,
}
def __A ( self : str , UpperCAmelCase : int ):
A_ = model_inputs.pop("target_size" )
A_ = model_inputs.pop("candidate_label" )
A_ = model_inputs.pop("is_last" )
A_ = self.model(**UpperCAmelCase )
A_ = {"target_size": target_size, "candidate_label": candidate_label, "is_last": is_last, **outputs}
return model_outputs
def __A ( self : Dict , UpperCAmelCase : Any , UpperCAmelCase : Optional[int]=0.1 , UpperCAmelCase : Optional[int]=None ):
A_ = []
for model_output in model_outputs:
A_ = model_output["candidate_label"]
A_ = BaseModelOutput(UpperCAmelCase )
A_ = self.image_processor.post_process_object_detection(
outputs=UpperCAmelCase , threshold=UpperCAmelCase , target_sizes=model_output["target_size"] )[0]
for index in outputs["scores"].nonzero():
A_ = outputs["scores"][index].item()
A_ = self._get_bounding_box(outputs["boxes"][index][0] )
A_ = {"score": score, "label": label, "box": box}
results.append(UpperCAmelCase )
A_ = sorted(UpperCAmelCase , key=lambda UpperCAmelCase : x["score"] , reverse=UpperCAmelCase )
if top_k:
A_ = results[:top_k]
return results
def __A ( self : List[str] , UpperCAmelCase : "torch.Tensor" ):
if self.framework != "pt":
raise ValueError("The ZeroShotObjectDetectionPipeline is only available in PyTorch." )
A_ , A_ , A_ , A_ = box.int().tolist()
A_ = {
"xmin": xmin,
"ymin": ymin,
"xmax": xmax,
"ymax": ymax,
}
return bbox | 86 | 1 |
import unittest
import numpy as np
from transformers.testing_utils import require_torch, require_vision
from transformers.utils import is_torch_available, is_vision_available
from ...test_image_processing_common import ImageProcessingSavingTestMixin, prepare_image_inputs
if is_torch_available():
import torch
if is_vision_available():
from PIL import Image
from transformers import GLPNImageProcessor
class _a ( unittest.TestCase ):
"""simple docstring"""
def __init__( self : List[Any] , UpperCAmelCase : List[str] , UpperCAmelCase : Any=7 , UpperCAmelCase : Dict=3 , UpperCAmelCase : Optional[Any]=18 , UpperCAmelCase : List[str]=30 , UpperCAmelCase : Optional[Any]=400 , UpperCAmelCase : str=True , UpperCAmelCase : Optional[int]=32 , UpperCAmelCase : Dict=True , ):
A_ = parent
A_ = batch_size
A_ = num_channels
A_ = image_size
A_ = min_resolution
A_ = max_resolution
A_ = do_resize
A_ = size_divisor
A_ = do_rescale
def __A ( self : Tuple ):
return {
"do_resize": self.do_resize,
"size_divisor": self.size_divisor,
"do_rescale": self.do_rescale,
}
@require_torch
@require_vision
class _a ( snake_case_ , unittest.TestCase ):
"""simple docstring"""
_lowerCamelCase : Any = GLPNImageProcessor if is_vision_available() else None
def __A ( self : Any ):
A_ = GLPNImageProcessingTester(self )
@property
def __A ( self : List[str] ):
return self.image_processor_tester.prepare_image_processor_dict()
def __A ( self : Optional[Any] ):
A_ = self.image_processing_class(**self.image_processor_dict )
self.assertTrue(hasattr(UpperCAmelCase , "do_resize" ) )
self.assertTrue(hasattr(UpperCAmelCase , "size_divisor" ) )
self.assertTrue(hasattr(UpperCAmelCase , "resample" ) )
self.assertTrue(hasattr(UpperCAmelCase , "do_rescale" ) )
def __A ( self : List[str] ):
pass
def __A ( self : Optional[Any] ):
# Initialize image_processing
A_ = self.image_processing_class(**self.image_processor_dict )
# create random PIL images
A_ = prepare_image_inputs(self.image_processor_tester , equal_resolution=UpperCAmelCase )
for image in image_inputs:
self.assertIsInstance(UpperCAmelCase , Image.Image )
# Test not batched input (GLPNImageProcessor doesn't support batching)
A_ = image_processing(image_inputs[0] , return_tensors="pt" ).pixel_values
self.assertTrue(encoded_images.shape[-1] % self.image_processor_tester.size_divisor == 0 )
self.assertTrue(encoded_images.shape[-2] % self.image_processor_tester.size_divisor == 0 )
def __A ( self : str ):
# Initialize image_processing
A_ = self.image_processing_class(**self.image_processor_dict )
# create random numpy tensors
A_ = prepare_image_inputs(self.image_processor_tester , equal_resolution=UpperCAmelCase , numpify=UpperCAmelCase )
for image in image_inputs:
self.assertIsInstance(UpperCAmelCase , np.ndarray )
# Test not batched input (GLPNImageProcessor doesn't support batching)
A_ = image_processing(image_inputs[0] , return_tensors="pt" ).pixel_values
self.assertTrue(encoded_images.shape[-1] % self.image_processor_tester.size_divisor == 0 )
self.assertTrue(encoded_images.shape[-2] % self.image_processor_tester.size_divisor == 0 )
def __A ( self : str ):
# Initialize image_processing
A_ = self.image_processing_class(**self.image_processor_dict )
# create random PyTorch tensors
A_ = prepare_image_inputs(self.image_processor_tester , equal_resolution=UpperCAmelCase , torchify=UpperCAmelCase )
for image in image_inputs:
self.assertIsInstance(UpperCAmelCase , torch.Tensor )
# Test not batched input (GLPNImageProcessor doesn't support batching)
A_ = image_processing(image_inputs[0] , return_tensors="pt" ).pixel_values
self.assertTrue(encoded_images.shape[-1] % self.image_processor_tester.size_divisor == 0 )
self.assertTrue(encoded_images.shape[-2] % self.image_processor_tester.size_divisor == 0 ) | 86 |
import argparse
import torch
# Step 1. clone https://github.com/microsoft/unilm
# Step 2. git checkout to https://github.com/microsoft/unilm/commit/b94ec76c36f02fb2b0bf0dcb0b8554a2185173cd
# Step 3. cd unilm
# Step 4. ln -s $(realpath wavlm/modules.py) ./ # create simlink
# import classes
from unilm.wavlm.WavLM import WavLM as WavLMOrig
from unilm.wavlm.WavLM import WavLMConfig as WavLMConfigOrig
from transformers import WavLMConfig, WavLMModel, logging
logging.set_verbosity_info()
__a :Any = logging.get_logger(__name__)
__a :int = {
'post_extract_proj': 'feature_projection.projection',
'encoder.pos_conv.0': 'encoder.pos_conv_embed.conv',
'self_attn.k_proj': 'encoder.layers.*.attention.k_proj',
'self_attn.v_proj': 'encoder.layers.*.attention.v_proj',
'self_attn.q_proj': 'encoder.layers.*.attention.q_proj',
'self_attn.out_proj': 'encoder.layers.*.attention.out_proj',
'self_attn.grep_linear': 'encoder.layers.*.attention.gru_rel_pos_linear',
'self_attn.relative_attention_bias': 'encoder.layers.*.attention.rel_attn_embed',
'self_attn.grep_a': 'encoder.layers.*.attention.gru_rel_pos_const',
'self_attn_layer_norm': 'encoder.layers.*.layer_norm',
'fc1': 'encoder.layers.*.feed_forward.intermediate_dense',
'fc2': 'encoder.layers.*.feed_forward.output_dense',
'final_layer_norm': 'encoder.layers.*.final_layer_norm',
'encoder.layer_norm': 'encoder.layer_norm',
'w2v_model.layer_norm': 'feature_projection.layer_norm',
'quantizer.weight_proj': 'quantizer.weight_proj',
'quantizer.vars': 'quantizer.codevectors',
'project_q': 'project_q',
'final_proj': 'project_hid',
'w2v_encoder.proj': 'ctc_proj',
'mask_emb': 'masked_spec_embed',
}
__a :Tuple = [
'ctc_proj',
'quantizer.weight_proj',
'quantizer.codevectors',
'project_q',
'project_hid',
]
def __snake_case ( __UpperCamelCase : List[str] ,__UpperCamelCase : List[Any] ,__UpperCamelCase : Dict ,__UpperCamelCase : Any ,__UpperCamelCase : Optional[int] ):
"""simple docstring"""
for attribute in key.split("." ):
A_ = getattr(__UpperCamelCase ,__UpperCamelCase )
if weight_type is not None:
A_ = getattr(__UpperCamelCase ,__UpperCamelCase ).shape
else:
A_ = hf_pointer.shape
assert hf_shape == value.shape, (
f'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
f''' {value.shape} for {full_name}'''
)
if weight_type == "weight":
A_ = value
elif weight_type == "weight_g":
A_ = value
elif weight_type == "weight_v":
A_ = value
elif weight_type == "bias":
A_ = value
else:
A_ = value
logger.info(f'''{key + "." + weight_type if weight_type is not None else ""} was initialized from {full_name}.''' )
def __snake_case ( __UpperCamelCase : Optional[int] ,__UpperCamelCase : Optional[Any] ):
"""simple docstring"""
A_ = []
A_ = fairseq_model.state_dict()
A_ = hf_model.feature_extractor
for name, value in fairseq_dict.items():
A_ = False
if "conv_layers" in name:
load_conv_layer(
__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,hf_model.config.feat_extract_norm == "group" ,)
A_ = True
else:
for key, mapped_key in MAPPING.items():
if key in name or key.split("w2v_model." )[-1] == name.split("." )[0]:
A_ = True
if "*" in mapped_key:
A_ = name.split(__UpperCamelCase )[0].split("." )[-2]
A_ = mapped_key.replace("*" ,__UpperCamelCase )
if "weight_g" in name:
A_ = "weight_g"
elif "weight_v" in name:
A_ = "weight_v"
elif "bias" in name and "relative_attention_bias" not in name:
A_ = "bias"
elif "weight" in name:
# TODO: don't match quantizer.weight_proj
A_ = "weight"
else:
A_ = None
set_recursively(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
continue
if not is_used:
unused_weights.append(__UpperCamelCase )
logger.warning(f'''Unused weights: {unused_weights}''' )
def __snake_case ( __UpperCamelCase : Optional[int] ,__UpperCamelCase : Dict ,__UpperCamelCase : List[str] ,__UpperCamelCase : Union[str, Any] ,__UpperCamelCase : Optional[int] ):
"""simple docstring"""
A_ = full_name.split("conv_layers." )[-1]
A_ = name.split("." )
A_ = int(items[0] )
A_ = int(items[1] )
if type_id == 0:
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.bias.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape, (
f'''{full_name} has size {value.shape}, but {feature_extractor[layer_id].layer_norm.bias.data.shape} was'''
" found."
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor[layer_id].layer_norm.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(__UpperCamelCase )
@torch.no_grad()
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : str ,__UpperCamelCase : int=None ):
"""simple docstring"""
A_ = torch.load(__UpperCamelCase )
A_ = WavLMConfigOrig(checkpoint["cfg"] )
A_ = WavLMOrig(__UpperCamelCase )
model.load_state_dict(checkpoint["model"] )
model.eval()
if config_path is not None:
A_ = WavLMConfig.from_pretrained(__UpperCamelCase )
else:
A_ = WavLMConfig()
A_ = WavLMModel(__UpperCamelCase )
recursively_load_weights(__UpperCamelCase ,__UpperCamelCase )
hf_wavlm.save_pretrained(__UpperCamelCase )
if __name__ == "__main__":
__a :List[Any] = argparse.ArgumentParser()
parser.add_argument('--pytorch_dump_folder_path', default=None, type=str, help='Path to the output PyTorch model.')
parser.add_argument('--checkpoint_path', default=None, type=str, help='Path to fairseq checkpoint')
parser.add_argument('--config_path', default=None, type=str, help='Path to hf config.json of model to convert')
__a :Optional[int] = parser.parse_args()
convert_wavlm_checkpoint(args.checkpoint_path, args.pytorch_dump_folder_path, args.config_path) | 86 | 1 |
from timeit import timeit
def __snake_case ( __UpperCamelCase : int ):
"""simple docstring"""
if number < 0:
raise ValueError("the value of input must not be negative" )
A_ = 0
while number:
number &= number - 1
result += 1
return result
def __snake_case ( __UpperCamelCase : int ):
"""simple docstring"""
if number < 0:
raise ValueError("the value of input must not be negative" )
A_ = 0
while number:
if number % 2 == 1:
result += 1
number >>= 1
return result
def __snake_case ( ):
"""simple docstring"""
def do_benchmark(__UpperCamelCase : int ) -> None:
A_ = "import __main__ as z"
print(f'''Benchmark when {number = }:''' )
print(f'''{get_set_bits_count_using_modulo_operator(__UpperCamelCase ) = }''' )
A_ = timeit("z.get_set_bits_count_using_modulo_operator(25)" ,setup=__UpperCamelCase )
print(f'''timeit() runs in {timing} seconds''' )
print(f'''{get_set_bits_count_using_brian_kernighans_algorithm(__UpperCamelCase ) = }''' )
A_ = timeit(
"z.get_set_bits_count_using_brian_kernighans_algorithm(25)" ,setup=__UpperCamelCase ,)
print(f'''timeit() runs in {timing} seconds''' )
for number in (25, 37, 58, 0):
do_benchmark(__UpperCamelCase )
print()
if __name__ == "__main__":
import doctest
doctest.testmod()
benchmark() | 86 |
def __snake_case ( __UpperCamelCase : list ,__UpperCamelCase : int = 0 ):
"""simple docstring"""
A_ = length or len(__UpperCamelCase )
A_ = False
for i in range(length - 1 ):
if list_data[i] > list_data[i + 1]:
A_ , A_ = list_data[i + 1], list_data[i]
A_ = True
return list_data if not swapped else bubble_sort(__UpperCamelCase ,length - 1 )
if __name__ == "__main__":
import doctest
doctest.testmod() | 86 | 1 |
from typing import TYPE_CHECKING
from ...utils import (
OptionalDependencyNotAvailable,
_LazyModule,
is_torch_available,
)
__a :List[Any] = {
'configuration_swiftformer': [
'SWIFTFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP',
'SwiftFormerConfig',
'SwiftFormerOnnxConfig',
]
}
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Union[str, Any] = [
'SWIFTFORMER_PRETRAINED_MODEL_ARCHIVE_LIST',
'SwiftFormerForImageClassification',
'SwiftFormerModel',
'SwiftFormerPreTrainedModel',
]
if TYPE_CHECKING:
from .configuration_swiftformer import (
SWIFTFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP,
SwiftFormerConfig,
SwiftFormerOnnxConfig,
)
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_swiftformer import (
SWIFTFORMER_PRETRAINED_MODEL_ARCHIVE_LIST,
SwiftFormerForImageClassification,
SwiftFormerModel,
SwiftFormerPreTrainedModel,
)
else:
import sys
__a :List[str] = _LazyModule(__name__, globals()['__file__'], _import_structure, module_spec=__spec__) | 86 |
import pickle
import unittest
import torch
from accelerate import Accelerator
from accelerate.state import AcceleratorState
from accelerate.test_utils import require_cpu
@require_cpu
class _a ( unittest.TestCase ):
"""simple docstring"""
def __A ( self : List[str] ):
A_ = torch.nn.Linear(10 , 10 )
A_ = torch.optim.SGD(model.parameters() , 0.1 )
A_ = Accelerator()
A_ = accelerator.prepare(UpperCAmelCase )
try:
pickle.loads(pickle.dumps(UpperCAmelCase ) )
except Exception as e:
self.fail(f'''Accelerated optimizer pickling failed with {e}''' )
AcceleratorState._reset_state() | 86 | 1 |
__a :Union[str, Any] = '\n# Transformers installation\n! pip install transformers datasets\n# To install from source instead of the last release, comment the command above and uncomment the following one.\n# ! pip install git+https://github.com/huggingface/transformers.git\n'
__a :List[str] = [{'type': 'code', 'content': INSTALL_CONTENT}]
__a :Dict = {
'{processor_class}': 'FakeProcessorClass',
'{model_class}': 'FakeModelClass',
'{object_class}': 'FakeObjectClass',
} | 86 |
import argparse
import json
import os
import fairseq
import torch
from torch import nn
from transformers import (
SpeechaTextaConfig,
SpeechaTextaForCausalLM,
SpeechaTextaTokenizer,
SpeechEncoderDecoderConfig,
SpeechEncoderDecoderModel,
WavaVecaConfig,
WavaVecaFeatureExtractor,
WavaVecaModel,
logging,
)
logging.set_verbosity_info()
__a :List[str] = logging.get_logger(__name__)
__a :Optional[int] = {
'post_extract_proj': 'feature_projection.projection',
'encoder.pos_conv.0': 'encoder.pos_conv_embed.conv',
'self_attn.k_proj': 'encoder.layers.*.attention.k_proj',
'self_attn.v_proj': 'encoder.layers.*.attention.v_proj',
'self_attn.q_proj': 'encoder.layers.*.attention.q_proj',
'self_attn.out_proj': 'encoder.layers.*.attention.out_proj',
'self_attn_layer_norm': 'encoder.layers.*.layer_norm',
'fc1': 'encoder.layers.*.feed_forward.intermediate_dense',
'fc2': 'encoder.layers.*.feed_forward.output_dense',
'final_layer_norm': 'encoder.layers.*.final_layer_norm',
'encoder.layer_norm': 'encoder.layer_norm',
'w2v_model.layer_norm': 'feature_projection.layer_norm',
'quantizer.weight_proj': 'quantizer.weight_proj',
'quantizer.vars': 'quantizer.codevectors',
'project_q': 'project_q',
'final_proj': 'project_hid',
'w2v_encoder.proj': 'lm_head',
'mask_emb': 'masked_spec_embed',
}
__a :Any = [
'lm_head',
'quantizer.weight_proj',
'quantizer.codevectors',
'project_q',
'project_hid',
]
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : Optional[Any] ,__UpperCamelCase : Optional[int] ,__UpperCamelCase : List[str] ,__UpperCamelCase : Union[str, Any] ):
"""simple docstring"""
for attribute in key.split("." ):
A_ = getattr(__UpperCamelCase ,__UpperCamelCase )
if weight_type is not None:
A_ = getattr(__UpperCamelCase ,__UpperCamelCase ).shape
else:
A_ = hf_pointer.shape
assert hf_shape == value.shape, (
f'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
f''' {value.shape} for {full_name}'''
)
if weight_type == "weight":
A_ = value
elif weight_type == "weight_g":
A_ = value
elif weight_type == "weight_v":
A_ = value
elif weight_type == "bias":
A_ = value
else:
A_ = value
logger.info(f'''{key + "." + weight_type if weight_type is not None else ""} was initialized from {full_name}.''' )
def __snake_case ( __UpperCamelCase : Union[str, Any] ,__UpperCamelCase : Dict ):
"""simple docstring"""
A_ = []
A_ = fairseq_model.state_dict()
A_ = hf_model.feature_extractor
# if encoder has different dim to decoder -> use proj_weight
A_ = None
for name, value in fairseq_dict.items():
A_ = False
if "conv_layers" in name:
load_conv_layer(
__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,hf_model.config.feat_extract_norm == "group" ,)
A_ = True
elif name.split("." )[0] == "proj":
A_ = fairseq_model.proj
A_ = True
else:
for key, mapped_key in MAPPING.items():
if key in name or key.split("w2v_model." )[-1] == name.split("." )[0]:
A_ = True
if "*" in mapped_key:
A_ = name.split(__UpperCamelCase )[0].split("." )[-2]
A_ = mapped_key.replace("*" ,__UpperCamelCase )
if "weight_g" in name:
A_ = "weight_g"
elif "weight_v" in name:
A_ = "weight_v"
elif "bias" in name:
A_ = "bias"
elif "weight" in name:
A_ = "weight"
else:
A_ = None
set_recursively(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
continue
if not is_used:
unused_weights.append(__UpperCamelCase )
logger.warning(f'''Unused weights: {unused_weights}''' )
return proj_weight
def __snake_case ( __UpperCamelCase : str ,__UpperCamelCase : Optional[Any] ,__UpperCamelCase : int ,__UpperCamelCase : int ,__UpperCamelCase : Any ):
"""simple docstring"""
A_ = full_name.split("conv_layers." )[-1]
A_ = name.split("." )
A_ = int(items[0] )
A_ = int(items[1] )
if type_id == 0:
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.bias.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape, (
f'''{full_name} has size {value.shape}, but {feature_extractor[layer_id].layer_norm.bias.data.shape} was'''
" found."
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor[layer_id].layer_norm.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(__UpperCamelCase )
def __snake_case ( __UpperCamelCase : Optional[Any] ):
"""simple docstring"""
A_ , A_ = emb.weight.shape
A_ = nn.Linear(__UpperCamelCase ,__UpperCamelCase ,bias=__UpperCamelCase )
A_ = emb.weight.data
return lin_layer
def __snake_case ( __UpperCamelCase : Tuple ):
"""simple docstring"""
with open(__UpperCamelCase ,"r" ,encoding="utf-8" ) as f:
A_ = f.readlines()
A_ = [line.split(" " )[0] for line in lines]
A_ = len(__UpperCamelCase )
A_ = {
"<s>": 0,
"<pad>": 1,
"</s>": 2,
"<unk>": 3,
}
vocab_dict.update(dict(zip(__UpperCamelCase ,range(4 ,num_words + 4 ) ) ) )
return vocab_dict
@torch.no_grad()
def __snake_case ( __UpperCamelCase : List[str] ,__UpperCamelCase : Any ,__UpperCamelCase : List[Any] ,__UpperCamelCase : Union[str, Any] ,__UpperCamelCase : List[str] ,__UpperCamelCase : Optional[Any] ,__UpperCamelCase : Dict ,):
"""simple docstring"""
A_ = WavaVecaConfig.from_pretrained(__UpperCamelCase )
A_ = SpeechaTextaConfig.from_pretrained(
__UpperCamelCase ,vocab_size=__UpperCamelCase ,decoder_layers=__UpperCamelCase ,do_stable_layer_norm=__UpperCamelCase )
A_ = WavaVecaFeatureExtractor(
feature_size=1 ,sampling_rate=1_6000 ,padding_value=0 ,do_normalize=__UpperCamelCase ,return_attention_mask=__UpperCamelCase ,)
A_ , A_ , A_ = fairseq.checkpoint_utils.load_model_ensemble_and_task(
[checkpoint_path] ,arg_overrides={"data": "/".join(dict_path.split("/" )[:-1] )} )
A_ = model[0].eval()
# set weights for wav2vec2 encoder
A_ = WavaVecaModel(__UpperCamelCase )
A_ = recursively_load_weights_wavaveca(model.encoder ,__UpperCamelCase )
A_ = SpeechaTextaForCausalLM(__UpperCamelCase )
A_ , A_ = hf_decoder.model.decoder.load_state_dict(model.decoder.state_dict() ,strict=__UpperCamelCase )
# set output linear layer
unexpected_keys.remove("embed_out" )
A_ = nn.Parameter(model.decoder.embed_out.detach() )
# layer norm is init to identity matrix so leaving it is fine
logger.warning(f'''The following keys are missing when loading the decoder weights: {missing_keys}''' )
logger.warning(f'''The following keys are unexpected when loading the decoder weights: {unexpected_keys}''' )
A_ = SpeechEncoderDecoderModel(encoder=__UpperCamelCase ,decoder=__UpperCamelCase )
A_ = False
# add projection layer
A_ = nn.Parameter(projection_layer.weight )
A_ = nn.Parameter(projection_layer.bias )
A_ = create_vocab_dict(__UpperCamelCase )
with open(os.path.join(__UpperCamelCase ,"vocab.json" ) ,"w" ) as fp:
json.dump(__UpperCamelCase ,__UpperCamelCase )
A_ = SpeechaTextaTokenizer(os.path.join(__UpperCamelCase ,"vocab.json" ) )
tokenizer.save_pretrained(__UpperCamelCase )
A_ = hf_wavavec.config.to_dict()
A_ = tokenizer.pad_token_id
A_ = tokenizer.bos_token_id
A_ = tokenizer.eos_token_id
A_ = "speech_to_text_2"
A_ = "wav2vec2"
A_ = SpeechEncoderDecoderConfig.from_dict(__UpperCamelCase )
hf_wavavec.save_pretrained(__UpperCamelCase )
feature_extractor.save_pretrained(__UpperCamelCase )
if __name__ == "__main__":
__a :int = argparse.ArgumentParser()
parser.add_argument('--pytorch_dump_folder_path', default=None, type=str, help='Path to the output PyTorch model.')
parser.add_argument('--checkpoint_path', default=None, type=str, help='Path to fairseq checkpoint')
parser.add_argument('--dict_path', default=None, type=str, help='Path to dict of fine-tuned model')
parser.add_argument(
'--encoder_config_path',
default='facebook/wav2vec2-large-lv60',
type=str,
help='Path to hf encoder wav2vec2 checkpoint config',
)
parser.add_argument(
'--decoder_config_path',
default='facebook/s2t-small-mustc-en-fr-st',
type=str,
help='Path to hf decoder s2t checkpoint config',
)
parser.add_argument('--vocab_size', default=1_0224, type=int, help='Vocab size of decoder')
parser.add_argument('--num_decoder_layers', default=7, type=int, help='Number of decoder layers')
__a :Tuple = parser.parse_args()
convert_wavaveca_checkpoint(
args.checkpoint_path,
args.pytorch_dump_folder_path,
args.dict_path,
encoder_config_path=args.encoder_config_path,
decoder_config_path=args.decoder_config_path,
vocab_size=args.vocab_size,
num_decoder_layers=args.num_decoder_layers,
) | 86 | 1 |
import inspect
import unittest
from transformers import ViTConfig
from transformers.testing_utils import (
require_accelerate,
require_torch,
require_torch_gpu,
require_vision,
slow,
torch_device,
)
from transformers.utils import cached_property, is_torch_available, is_vision_available
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from torch import nn
from transformers import ViTForImageClassification, ViTForMaskedImageModeling, ViTModel
from transformers.models.vit.modeling_vit import VIT_PRETRAINED_MODEL_ARCHIVE_LIST
if is_vision_available():
from PIL import Image
from transformers import ViTImageProcessor
class _a :
"""simple docstring"""
def __init__( self : Optional[int] , UpperCAmelCase : Any , UpperCAmelCase : Any=13 , UpperCAmelCase : Any=30 , UpperCAmelCase : int=2 , UpperCAmelCase : Union[str, Any]=3 , UpperCAmelCase : List[Any]=True , UpperCAmelCase : str=True , UpperCAmelCase : Dict=32 , UpperCAmelCase : str=5 , UpperCAmelCase : Any=4 , UpperCAmelCase : str=37 , UpperCAmelCase : List[Any]="gelu" , UpperCAmelCase : List[Any]=0.1 , UpperCAmelCase : int=0.1 , UpperCAmelCase : Optional[Any]=10 , UpperCAmelCase : List[Any]=0.02 , UpperCAmelCase : Union[str, Any]=None , UpperCAmelCase : Tuple=2 , ):
A_ = parent
A_ = batch_size
A_ = image_size
A_ = patch_size
A_ = num_channels
A_ = is_training
A_ = use_labels
A_ = hidden_size
A_ = num_hidden_layers
A_ = num_attention_heads
A_ = intermediate_size
A_ = hidden_act
A_ = hidden_dropout_prob
A_ = attention_probs_dropout_prob
A_ = type_sequence_label_size
A_ = initializer_range
A_ = scope
A_ = encoder_stride
# in ViT, the seq length equals the number of patches + 1 (we add 1 for the [CLS] token)
A_ = (image_size // patch_size) ** 2
A_ = num_patches + 1
def __A ( self : Union[str, Any] ):
A_ = floats_tensor([self.batch_size, self.num_channels, self.image_size, self.image_size] )
A_ = None
if self.use_labels:
A_ = ids_tensor([self.batch_size] , self.type_sequence_label_size )
A_ = self.get_config()
return config, pixel_values, labels
def __A ( self : Optional[Any] ):
return ViTConfig(
image_size=self.image_size , patch_size=self.patch_size , num_channels=self.num_channels , hidden_size=self.hidden_size , num_hidden_layers=self.num_hidden_layers , num_attention_heads=self.num_attention_heads , intermediate_size=self.intermediate_size , hidden_act=self.hidden_act , hidden_dropout_prob=self.hidden_dropout_prob , attention_probs_dropout_prob=self.attention_probs_dropout_prob , is_decoder=UpperCAmelCase , initializer_range=self.initializer_range , encoder_stride=self.encoder_stride , )
def __A ( self : Tuple , UpperCAmelCase : str , UpperCAmelCase : Any , UpperCAmelCase : Any ):
A_ = ViTModel(config=UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
A_ = model(UpperCAmelCase )
self.parent.assertEqual(result.last_hidden_state.shape , (self.batch_size, self.seq_length, self.hidden_size) )
def __A ( self : Tuple , UpperCAmelCase : Union[str, Any] , UpperCAmelCase : Any , UpperCAmelCase : Optional[Any] ):
A_ = ViTForMaskedImageModeling(config=UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
A_ = model(UpperCAmelCase )
self.parent.assertEqual(
result.reconstruction.shape , (self.batch_size, self.num_channels, self.image_size, self.image_size) )
# test greyscale images
A_ = 1
A_ = ViTForMaskedImageModeling(UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
A_ = floats_tensor([self.batch_size, 1, self.image_size, self.image_size] )
A_ = model(UpperCAmelCase )
self.parent.assertEqual(result.reconstruction.shape , (self.batch_size, 1, self.image_size, self.image_size) )
def __A ( self : str , UpperCAmelCase : int , UpperCAmelCase : Any , UpperCAmelCase : Any ):
A_ = self.type_sequence_label_size
A_ = ViTForImageClassification(UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
A_ = model(UpperCAmelCase , labels=UpperCAmelCase )
self.parent.assertEqual(result.logits.shape , (self.batch_size, self.type_sequence_label_size) )
# test greyscale images
A_ = 1
A_ = ViTForImageClassification(UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
A_ = floats_tensor([self.batch_size, 1, self.image_size, self.image_size] )
A_ = model(UpperCAmelCase )
self.parent.assertEqual(result.logits.shape , (self.batch_size, self.type_sequence_label_size) )
def __A ( self : Optional[Any] ):
A_ = self.prepare_config_and_inputs()
(
(
A_
) , (
A_
) , (
A_
) ,
) = config_and_inputs
A_ = {"pixel_values": pixel_values}
return config, inputs_dict
@require_torch
class _a ( snake_case_ , snake_case_ , unittest.TestCase ):
"""simple docstring"""
_lowerCamelCase : List[str] = (
(
ViTModel,
ViTForImageClassification,
ViTForMaskedImageModeling,
)
if is_torch_available()
else ()
)
_lowerCamelCase : Optional[Any] = (
{'feature-extraction': ViTModel, 'image-classification': ViTForImageClassification}
if is_torch_available()
else {}
)
_lowerCamelCase : List[Any] = True
_lowerCamelCase : List[str] = False
_lowerCamelCase : str = False
_lowerCamelCase : Tuple = False
def __A ( self : Optional[Any] ):
A_ = ViTModelTester(self )
A_ = ConfigTester(self , config_class=UpperCAmelCase , has_text_modality=UpperCAmelCase , hidden_size=37 )
def __A ( self : Optional[Any] ):
self.config_tester.run_common_tests()
@unittest.skip(reason="ViT does not use inputs_embeds" )
def __A ( self : Tuple ):
pass
def __A ( self : int ):
A_ , A_ = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
A_ = model_class(UpperCAmelCase )
self.assertIsInstance(model.get_input_embeddings() , (nn.Module) )
A_ = model.get_output_embeddings()
self.assertTrue(x is None or isinstance(UpperCAmelCase , nn.Linear ) )
def __A ( self : Tuple ):
A_ , A_ = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
A_ = model_class(UpperCAmelCase )
A_ = inspect.signature(model.forward )
# signature.parameters is an OrderedDict => so arg_names order is deterministic
A_ = [*signature.parameters.keys()]
A_ = ["pixel_values"]
self.assertListEqual(arg_names[:1] , UpperCAmelCase )
def __A ( self : Tuple ):
A_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*UpperCAmelCase )
def __A ( self : int ):
A_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_masked_image_modeling(*UpperCAmelCase )
def __A ( self : str ):
A_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_image_classification(*UpperCAmelCase )
@slow
def __A ( self : Tuple ):
for model_name in VIT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
A_ = ViTModel.from_pretrained(UpperCAmelCase )
self.assertIsNotNone(UpperCAmelCase )
def __snake_case ( ):
"""simple docstring"""
A_ = Image.open("./tests/fixtures/tests_samples/COCO/000000039769.png" )
return image
@require_torch
@require_vision
class _a ( unittest.TestCase ):
"""simple docstring"""
@cached_property
def __A ( self : List[Any] ):
return ViTImageProcessor.from_pretrained("google/vit-base-patch16-224" ) if is_vision_available() else None
@slow
def __A ( self : str ):
A_ = ViTForImageClassification.from_pretrained("google/vit-base-patch16-224" ).to(UpperCAmelCase )
A_ = self.default_image_processor
A_ = prepare_img()
A_ = image_processor(images=UpperCAmelCase , return_tensors="pt" ).to(UpperCAmelCase )
# forward pass
with torch.no_grad():
A_ = model(**UpperCAmelCase )
# verify the logits
A_ = torch.Size((1, 1000) )
self.assertEqual(outputs.logits.shape , UpperCAmelCase )
A_ = torch.tensor([-0.2_744, 0.8_215, -0.0_836] ).to(UpperCAmelCase )
self.assertTrue(torch.allclose(outputs.logits[0, :3] , UpperCAmelCase , atol=1E-4 ) )
@slow
def __A ( self : Any ):
# ViT models have an `interpolate_pos_encoding` argument in their forward method,
# allowing to interpolate the pre-trained position embeddings in order to use
# the model on higher resolutions. The DINO model by Facebook AI leverages this
# to visualize self-attention on higher resolution images.
A_ = ViTModel.from_pretrained("facebook/dino-vits8" ).to(UpperCAmelCase )
A_ = ViTImageProcessor.from_pretrained("facebook/dino-vits8" , size=480 )
A_ = prepare_img()
A_ = image_processor(images=UpperCAmelCase , return_tensors="pt" )
A_ = inputs.pixel_values.to(UpperCAmelCase )
# forward pass
with torch.no_grad():
A_ = model(UpperCAmelCase , interpolate_pos_encoding=UpperCAmelCase )
# verify the logits
A_ = torch.Size((1, 3601, 384) )
self.assertEqual(outputs.last_hidden_state.shape , UpperCAmelCase )
A_ = torch.tensor(
[[4.2_340, 4.3_906, -6.6_692], [4.5_463, 1.8_928, -6.7_257], [4.4_429, 0.8_496, -5.8_585]] ).to(UpperCAmelCase )
self.assertTrue(torch.allclose(outputs.last_hidden_state[0, :3, :3] , UpperCAmelCase , atol=1E-4 ) )
@slow
@require_accelerate
@require_torch_gpu
def __A ( self : Optional[int] ):
A_ = ViTModel.from_pretrained("facebook/dino-vits8" , torch_dtype=torch.floataa , device_map="auto" )
A_ = self.default_image_processor
A_ = prepare_img()
A_ = image_processor(images=UpperCAmelCase , return_tensors="pt" )
A_ = inputs.pixel_values.to(UpperCAmelCase )
# forward pass to make sure inference works in fp16
with torch.no_grad():
A_ = model(UpperCAmelCase ) | 86 |
from typing import Any, Dict, List, Union
from ..utils import add_end_docstrings, is_torch_available, is_vision_available, logging, requires_backends
from .base import PIPELINE_INIT_ARGS, Pipeline
if is_vision_available():
from ..image_utils import load_image
if is_torch_available():
import torch
from ..models.auto.modeling_auto import MODEL_FOR_OBJECT_DETECTION_MAPPING, MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING
__a :str = logging.get_logger(__name__)
__a :Any = Dict[str, Any]
__a :int = List[Prediction]
@add_end_docstrings(snake_case_ )
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Tuple , *UpperCAmelCase : Optional[int] , **UpperCAmelCase : Optional[Any] ):
super().__init__(*UpperCAmelCase , **UpperCAmelCase )
if self.framework == "tf":
raise ValueError(f'''The {self.__class__} is only available in PyTorch.''' )
requires_backends(self , "vision" )
self.check_model_type(
dict(MODEL_FOR_OBJECT_DETECTION_MAPPING.items() + MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING.items() ) )
def __A ( self : str , **UpperCAmelCase : str ):
A_ = {}
if "threshold" in kwargs:
A_ = kwargs["threshold"]
return {}, {}, postprocess_kwargs
def __call__( self : Union[str, Any] , *UpperCAmelCase : Union[str, Any] , **UpperCAmelCase : Optional[Any] ):
return super().__call__(*UpperCAmelCase , **UpperCAmelCase )
def __A ( self : str , UpperCAmelCase : Any ):
A_ = load_image(UpperCAmelCase )
A_ = torch.IntTensor([[image.height, image.width]] )
A_ = self.image_processor(images=[image] , return_tensors="pt" )
if self.tokenizer is not None:
A_ = self.tokenizer(text=inputs["words"] , boxes=inputs["boxes"] , return_tensors="pt" )
A_ = target_size
return inputs
def __A ( self : Optional[Any] , UpperCAmelCase : Optional[int] ):
A_ = model_inputs.pop("target_size" )
A_ = self.model(**UpperCAmelCase )
A_ = outputs.__class__({"target_size": target_size, **outputs} )
if self.tokenizer is not None:
A_ = model_inputs["bbox"]
return model_outputs
def __A ( self : str , UpperCAmelCase : Optional[int] , UpperCAmelCase : Union[str, Any]=0.9 ):
A_ = model_outputs["target_size"]
if self.tokenizer is not None:
# This is a LayoutLMForTokenClassification variant.
# The OCR got the boxes and the model classified the words.
A_ , A_ = target_size[0].tolist()
def unnormalize(UpperCAmelCase : Any ):
return self._get_bounding_box(
torch.Tensor(
[
(width * bbox[0] / 1000),
(height * bbox[1] / 1000),
(width * bbox[2] / 1000),
(height * bbox[3] / 1000),
] ) )
A_ , A_ = model_outputs["logits"].squeeze(0 ).softmax(dim=-1 ).max(dim=-1 )
A_ = [self.model.config.idalabel[prediction] for prediction in classes.tolist()]
A_ = [unnormalize(UpperCAmelCase ) for bbox in model_outputs["bbox"].squeeze(0 )]
A_ = ["score", "label", "box"]
A_ = [dict(zip(UpperCAmelCase , UpperCAmelCase ) ) for vals in zip(scores.tolist() , UpperCAmelCase , UpperCAmelCase ) if vals[0] > threshold]
else:
# This is a regular ForObjectDetectionModel
A_ = self.image_processor.post_process_object_detection(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase )
A_ = raw_annotations[0]
A_ = raw_annotation["scores"]
A_ = raw_annotation["labels"]
A_ = raw_annotation["boxes"]
A_ = scores.tolist()
A_ = [self.model.config.idalabel[label.item()] for label in labels]
A_ = [self._get_bounding_box(UpperCAmelCase ) for box in boxes]
# {"scores": [...], ...} --> [{"score":x, ...}, ...]
A_ = ["score", "label", "box"]
A_ = [
dict(zip(UpperCAmelCase , UpperCAmelCase ) )
for vals in zip(raw_annotation["scores"] , raw_annotation["labels"] , raw_annotation["boxes"] )
]
return annotation
def __A ( self : Tuple , UpperCAmelCase : "torch.Tensor" ):
if self.framework != "pt":
raise ValueError("The ObjectDetectionPipeline is only available in PyTorch." )
A_ , A_ , A_ , A_ = box.int().tolist()
A_ = {
"xmin": xmin,
"ymin": ymin,
"xmax": xmax,
"ymax": ymax,
}
return bbox | 86 | 1 |
import os
import tempfile
import unittest
from transformers import DistilBertConfig, is_torch_available
from transformers.testing_utils import require_torch, require_torch_gpu, slow, torch_device
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from transformers import (
DISTILBERT_PRETRAINED_MODEL_ARCHIVE_LIST,
DistilBertForMaskedLM,
DistilBertForMultipleChoice,
DistilBertForQuestionAnswering,
DistilBertForSequenceClassification,
DistilBertForTokenClassification,
DistilBertModel,
)
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Any , UpperCAmelCase : int , UpperCAmelCase : Optional[int]=13 , UpperCAmelCase : List[Any]=7 , UpperCAmelCase : Dict=True , UpperCAmelCase : Any=True , UpperCAmelCase : Dict=False , UpperCAmelCase : Union[str, Any]=True , UpperCAmelCase : Dict=99 , UpperCAmelCase : Tuple=32 , UpperCAmelCase : str=5 , UpperCAmelCase : List[Any]=4 , UpperCAmelCase : Any=37 , UpperCAmelCase : str="gelu" , UpperCAmelCase : Any=0.1 , UpperCAmelCase : Optional[int]=0.1 , UpperCAmelCase : int=512 , UpperCAmelCase : Union[str, Any]=16 , UpperCAmelCase : Tuple=2 , UpperCAmelCase : List[Any]=0.02 , UpperCAmelCase : Any=3 , UpperCAmelCase : Any=4 , UpperCAmelCase : Any=None , ):
A_ = parent
A_ = batch_size
A_ = seq_length
A_ = is_training
A_ = use_input_mask
A_ = use_token_type_ids
A_ = use_labels
A_ = vocab_size
A_ = hidden_size
A_ = num_hidden_layers
A_ = num_attention_heads
A_ = intermediate_size
A_ = hidden_act
A_ = hidden_dropout_prob
A_ = attention_probs_dropout_prob
A_ = max_position_embeddings
A_ = type_vocab_size
A_ = type_sequence_label_size
A_ = initializer_range
A_ = num_labels
A_ = num_choices
A_ = scope
def __A ( self : Optional[int] ):
A_ = ids_tensor([self.batch_size, self.seq_length] , self.vocab_size )
A_ = None
if self.use_input_mask:
A_ = random_attention_mask([self.batch_size, self.seq_length] )
A_ = None
A_ = None
A_ = None
if self.use_labels:
A_ = ids_tensor([self.batch_size] , self.type_sequence_label_size )
A_ = ids_tensor([self.batch_size, self.seq_length] , self.num_labels )
A_ = ids_tensor([self.batch_size] , self.num_choices )
A_ = self.get_config()
return config, input_ids, input_mask, sequence_labels, token_labels, choice_labels
def __A ( self : Optional[Any] ):
return DistilBertConfig(
vocab_size=self.vocab_size , dim=self.hidden_size , n_layers=self.num_hidden_layers , n_heads=self.num_attention_heads , hidden_dim=self.intermediate_size , hidden_act=self.hidden_act , dropout=self.hidden_dropout_prob , attention_dropout=self.attention_probs_dropout_prob , max_position_embeddings=self.max_position_embeddings , initializer_range=self.initializer_range , )
def __A ( self : Union[str, Any] , UpperCAmelCase : str , UpperCAmelCase : int , UpperCAmelCase : int , UpperCAmelCase : Optional[int] , UpperCAmelCase : List[Any] , UpperCAmelCase : List[Any] ):
A_ = DistilBertModel(config=UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
A_ = model(UpperCAmelCase , UpperCAmelCase )
A_ = model(UpperCAmelCase )
self.parent.assertEqual(result.last_hidden_state.shape , (self.batch_size, self.seq_length, self.hidden_size) )
def __A ( self : Optional[int] , UpperCAmelCase : Union[str, Any] , UpperCAmelCase : List[str] , UpperCAmelCase : Optional[Any] , UpperCAmelCase : int , UpperCAmelCase : Any , UpperCAmelCase : List[Any] ):
A_ = DistilBertForMaskedLM(config=UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
A_ = model(UpperCAmelCase , attention_mask=UpperCAmelCase , labels=UpperCAmelCase )
self.parent.assertEqual(result.logits.shape , (self.batch_size, self.seq_length, self.vocab_size) )
def __A ( self : str , UpperCAmelCase : Dict , UpperCAmelCase : List[str] , UpperCAmelCase : str , UpperCAmelCase : int , UpperCAmelCase : int , UpperCAmelCase : Optional[int] ):
A_ = DistilBertForQuestionAnswering(config=UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
A_ = model(
UpperCAmelCase , attention_mask=UpperCAmelCase , start_positions=UpperCAmelCase , end_positions=UpperCAmelCase )
self.parent.assertEqual(result.start_logits.shape , (self.batch_size, self.seq_length) )
self.parent.assertEqual(result.end_logits.shape , (self.batch_size, self.seq_length) )
def __A ( self : Union[str, Any] , UpperCAmelCase : int , UpperCAmelCase : List[str] , UpperCAmelCase : Optional[Any] , UpperCAmelCase : Union[str, Any] , UpperCAmelCase : Dict , UpperCAmelCase : List[Any] ):
A_ = self.num_labels
A_ = DistilBertForSequenceClassification(UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
A_ = model(UpperCAmelCase , attention_mask=UpperCAmelCase , labels=UpperCAmelCase )
self.parent.assertEqual(result.logits.shape , (self.batch_size, self.num_labels) )
def __A ( self : int , UpperCAmelCase : Any , UpperCAmelCase : str , UpperCAmelCase : List[Any] , UpperCAmelCase : Optional[Any] , UpperCAmelCase : Optional[int] , UpperCAmelCase : Any ):
A_ = self.num_labels
A_ = DistilBertForTokenClassification(config=UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
A_ = model(UpperCAmelCase , attention_mask=UpperCAmelCase , labels=UpperCAmelCase )
self.parent.assertEqual(result.logits.shape , (self.batch_size, self.seq_length, self.num_labels) )
def __A ( self : Any , UpperCAmelCase : List[str] , UpperCAmelCase : Optional[Any] , UpperCAmelCase : Union[str, Any] , UpperCAmelCase : Any , UpperCAmelCase : List[Any] , UpperCAmelCase : Optional[int] ):
A_ = self.num_choices
A_ = DistilBertForMultipleChoice(config=UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
A_ = input_ids.unsqueeze(1 ).expand(-1 , self.num_choices , -1 ).contiguous()
A_ = input_mask.unsqueeze(1 ).expand(-1 , self.num_choices , -1 ).contiguous()
A_ = model(
UpperCAmelCase , attention_mask=UpperCAmelCase , labels=UpperCAmelCase , )
self.parent.assertEqual(result.logits.shape , (self.batch_size, self.num_choices) )
def __A ( self : str ):
A_ = self.prepare_config_and_inputs()
((A_) , (A_) , (A_) , (A_) , (A_) , (A_)) = config_and_inputs
A_ = {"input_ids": input_ids, "attention_mask": input_mask}
return config, inputs_dict
@require_torch
class _a ( snake_case_ , snake_case_ , unittest.TestCase ):
"""simple docstring"""
_lowerCamelCase : Optional[Any] = (
(
DistilBertModel,
DistilBertForMaskedLM,
DistilBertForMultipleChoice,
DistilBertForQuestionAnswering,
DistilBertForSequenceClassification,
DistilBertForTokenClassification,
)
if is_torch_available()
else None
)
_lowerCamelCase : List[Any] = (
{
'feature-extraction': DistilBertModel,
'fill-mask': DistilBertForMaskedLM,
'question-answering': DistilBertForQuestionAnswering,
'text-classification': DistilBertForSequenceClassification,
'token-classification': DistilBertForTokenClassification,
'zero-shot': DistilBertForSequenceClassification,
}
if is_torch_available()
else {}
)
_lowerCamelCase : Any = True
_lowerCamelCase : int = True
_lowerCamelCase : Union[str, Any] = True
_lowerCamelCase : str = True
def __A ( self : Union[str, Any] ):
A_ = DistilBertModelTester(self )
A_ = ConfigTester(self , config_class=UpperCAmelCase , dim=37 )
def __A ( self : Dict ):
self.config_tester.run_common_tests()
def __A ( self : Tuple ):
A_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_distilbert_model(*UpperCAmelCase )
def __A ( self : int ):
A_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_distilbert_for_masked_lm(*UpperCAmelCase )
def __A ( self : Dict ):
A_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_distilbert_for_question_answering(*UpperCAmelCase )
def __A ( self : Any ):
A_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_distilbert_for_sequence_classification(*UpperCAmelCase )
def __A ( self : int ):
A_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_distilbert_for_token_classification(*UpperCAmelCase )
def __A ( self : List[str] ):
A_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_distilbert_for_multiple_choice(*UpperCAmelCase )
@slow
def __A ( self : Optional[int] ):
for model_name in DISTILBERT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
A_ = DistilBertModel.from_pretrained(UpperCAmelCase )
self.assertIsNotNone(UpperCAmelCase )
@slow
@require_torch_gpu
def __A ( self : Optional[int] ):
A_ , A_ = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
# BertForMultipleChoice behaves incorrectly in JIT environments.
if model_class == DistilBertForMultipleChoice:
return
A_ = True
A_ = model_class(config=UpperCAmelCase )
A_ = self._prepare_for_class(UpperCAmelCase , UpperCAmelCase )
A_ = torch.jit.trace(
UpperCAmelCase , (inputs_dict["input_ids"].to("cpu" ), inputs_dict["attention_mask"].to("cpu" )) )
with tempfile.TemporaryDirectory() as tmp:
torch.jit.save(UpperCAmelCase , os.path.join(UpperCAmelCase , "traced_model.pt" ) )
A_ = torch.jit.load(os.path.join(UpperCAmelCase , "traced_model.pt" ) , map_location=UpperCAmelCase )
loaded(inputs_dict["input_ids"].to(UpperCAmelCase ) , inputs_dict["attention_mask"].to(UpperCAmelCase ) )
@require_torch
class _a ( unittest.TestCase ):
"""simple docstring"""
@slow
def __A ( self : Optional[int] ):
A_ = DistilBertModel.from_pretrained("distilbert-base-uncased" )
A_ = torch.tensor([[0, 345, 232, 328, 740, 140, 1695, 69, 6078, 1588, 2]] )
A_ = torch.tensor([[0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]] )
with torch.no_grad():
A_ = model(UpperCAmelCase , attention_mask=UpperCAmelCase )[0]
A_ = torch.Size((1, 11, 768) )
self.assertEqual(output.shape , UpperCAmelCase )
A_ = torch.tensor(
[[[-0.1_639, 0.3_299, 0.1_648], [-0.1_746, 0.3_289, 0.1_710], [-0.1_884, 0.3_357, 0.1_810]]] )
self.assertTrue(torch.allclose(output[:, 1:4, 1:4] , UpperCAmelCase , atol=1E-4 ) ) | 86 |
import inspect
from typing import List, Optional, Tuple, Union
import numpy as np
import PIL
import torch
import torch.utils.checkpoint
from ...models import UNetaDModel, VQModel
from ...schedulers import (
DDIMScheduler,
DPMSolverMultistepScheduler,
EulerAncestralDiscreteScheduler,
EulerDiscreteScheduler,
LMSDiscreteScheduler,
PNDMScheduler,
)
from ...utils import PIL_INTERPOLATION, randn_tensor
from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
def __snake_case ( __UpperCamelCase : Dict ):
"""simple docstring"""
A_ , A_ = image.size
A_ , A_ = (x - x % 32 for x in (w, h)) # resize to integer multiple of 32
A_ = image.resize((w, h) ,resample=PIL_INTERPOLATION["lanczos"] )
A_ = np.array(__UpperCamelCase ).astype(np.floataa ) / 255.0
A_ = image[None].transpose(0 ,3 ,1 ,2 )
A_ = torch.from_numpy(__UpperCamelCase )
return 2.0 * image - 1.0
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Union[str, Any] , UpperCAmelCase : VQModel , UpperCAmelCase : UNetaDModel , UpperCAmelCase : Union[
DDIMScheduler,
PNDMScheduler,
LMSDiscreteScheduler,
EulerDiscreteScheduler,
EulerAncestralDiscreteScheduler,
DPMSolverMultistepScheduler,
] , ):
super().__init__()
self.register_modules(vqvae=UpperCAmelCase , unet=UpperCAmelCase , scheduler=UpperCAmelCase )
@torch.no_grad()
def __call__( self : int , UpperCAmelCase : Union[torch.Tensor, PIL.Image.Image] = None , UpperCAmelCase : Optional[int] = 1 , UpperCAmelCase : Optional[int] = 100 , UpperCAmelCase : Optional[float] = 0.0 , UpperCAmelCase : Optional[Union[torch.Generator, List[torch.Generator]]] = None , UpperCAmelCase : Optional[str] = "pil" , UpperCAmelCase : bool = True , ):
if isinstance(UpperCAmelCase , PIL.Image.Image ):
A_ = 1
elif isinstance(UpperCAmelCase , torch.Tensor ):
A_ = image.shape[0]
else:
raise ValueError(f'''`image` has to be of type `PIL.Image.Image` or `torch.Tensor` but is {type(UpperCAmelCase )}''' )
if isinstance(UpperCAmelCase , PIL.Image.Image ):
A_ = preprocess(UpperCAmelCase )
A_ , A_ = image.shape[-2:]
# in_channels should be 6: 3 for latents, 3 for low resolution image
A_ = (batch_size, self.unet.config.in_channels // 2, height, width)
A_ = next(self.unet.parameters() ).dtype
A_ = randn_tensor(UpperCAmelCase , generator=UpperCAmelCase , device=self.device , dtype=UpperCAmelCase )
A_ = image.to(device=self.device , dtype=UpperCAmelCase )
# set timesteps and move to the correct device
self.scheduler.set_timesteps(UpperCAmelCase , device=self.device )
A_ = self.scheduler.timesteps
# scale the initial noise by the standard deviation required by the scheduler
A_ = latents * self.scheduler.init_noise_sigma
# prepare extra kwargs for the scheduler step, since not all schedulers have the same signature.
# eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
# eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
# and should be between [0, 1]
A_ = "eta" in set(inspect.signature(self.scheduler.step ).parameters.keys() )
A_ = {}
if accepts_eta:
A_ = eta
for t in self.progress_bar(UpperCAmelCase ):
# concat latents and low resolution image in the channel dimension.
A_ = torch.cat([latents, image] , dim=1 )
A_ = self.scheduler.scale_model_input(UpperCAmelCase , UpperCAmelCase )
# predict the noise residual
A_ = self.unet(UpperCAmelCase , UpperCAmelCase ).sample
# compute the previous noisy sample x_t -> x_t-1
A_ = self.scheduler.step(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase , **UpperCAmelCase ).prev_sample
# decode the image latents with the VQVAE
A_ = self.vqvae.decode(UpperCAmelCase ).sample
A_ = torch.clamp(UpperCAmelCase , -1.0 , 1.0 )
A_ = image / 2 + 0.5
A_ = image.cpu().permute(0 , 2 , 3 , 1 ).numpy()
if output_type == "pil":
A_ = self.numpy_to_pil(UpperCAmelCase )
if not return_dict:
return (image,)
return ImagePipelineOutput(images=UpperCAmelCase ) | 86 | 1 |
from typing import Dict, List, Optional, Union
import numpy as np
from ...image_processing_utils import BaseImageProcessor, BatchFeature, get_size_dict
from ...image_transforms import (
center_crop,
get_resize_output_image_size,
normalize,
rescale,
resize,
to_channel_dimension_format,
)
from ...image_utils import (
IMAGENET_DEFAULT_MEAN,
IMAGENET_DEFAULT_STD,
ChannelDimension,
ImageInput,
PILImageResampling,
is_batched,
to_numpy_array,
valid_images,
)
from ...utils import TensorType, logging
__a :Dict = logging.get_logger(__name__)
class _a ( snake_case_ ):
"""simple docstring"""
_lowerCamelCase : List[Any] = ['pixel_values']
def __init__( self : Dict , UpperCAmelCase : bool = True , UpperCAmelCase : Optional[Dict[str, int]] = None , UpperCAmelCase : PILImageResampling = PILImageResampling.BICUBIC , UpperCAmelCase : bool = True , UpperCAmelCase : bool = True , UpperCAmelCase : Union[int, float] = 1 / 255 , UpperCAmelCase : Dict[str, int] = None , UpperCAmelCase : bool = True , UpperCAmelCase : Optional[Union[float, List[float]]] = None , UpperCAmelCase : Optional[Union[float, List[float]]] = None , **UpperCAmelCase : Optional[int] , ):
super().__init__(**UpperCAmelCase )
A_ = size if size is not None else {"height": 224, "width": 224}
A_ = get_size_dict(UpperCAmelCase )
A_ = crop_size if crop_size is not None else {"height": 224, "width": 224}
A_ = get_size_dict(UpperCAmelCase , default_to_square=UpperCAmelCase , param_name="crop_size" )
A_ = do_resize
A_ = do_rescale
A_ = do_normalize
A_ = do_center_crop
A_ = crop_size
A_ = size
A_ = resample
A_ = rescale_factor
A_ = image_mean if image_mean is not None else IMAGENET_DEFAULT_MEAN
A_ = image_std if image_std is not None else IMAGENET_DEFAULT_STD
def __A ( self : Any , UpperCAmelCase : np.ndarray , UpperCAmelCase : Dict[str, int] , UpperCAmelCase : PILImageResampling = PILImageResampling.BILINEAR , UpperCAmelCase : Optional[Union[str, ChannelDimension]] = None , **UpperCAmelCase : Optional[int] , ):
A_ = get_size_dict(UpperCAmelCase )
if "shortest_edge" in size:
A_ = get_resize_output_image_size(UpperCAmelCase , size=size["shortest_edge"] , default_to_square=UpperCAmelCase )
# size = get_resize_output_image_size(image, size["shortest_edge"], size["longest_edge"])
elif "height" in size and "width" in size:
A_ = (size["height"], size["width"])
else:
raise ValueError(f'''Size must contain \'height\' and \'width\' keys or \'shortest_edge\' key. Got {size.keys()}''' )
return resize(UpperCAmelCase , size=UpperCAmelCase , resample=UpperCAmelCase , data_format=UpperCAmelCase , **UpperCAmelCase )
def __A ( self : str , UpperCAmelCase : np.ndarray , UpperCAmelCase : Dict[str, int] , UpperCAmelCase : Optional[Union[str, ChannelDimension]] = None , **UpperCAmelCase : Optional[Any] , ):
A_ = get_size_dict(UpperCAmelCase )
if "height" not in size or "width" not in size:
raise ValueError(f'''The `size` parameter must contain the keys (height, width). Got {size.keys()}''' )
return center_crop(UpperCAmelCase , size=(size["height"], size["width"]) , data_format=UpperCAmelCase , **UpperCAmelCase )
def __A ( self : str , UpperCAmelCase : np.ndarray , UpperCAmelCase : float , UpperCAmelCase : Optional[Union[str, ChannelDimension]] = None , **UpperCAmelCase : Any ):
return rescale(UpperCAmelCase , scale=UpperCAmelCase , data_format=UpperCAmelCase , **UpperCAmelCase )
def __A ( self : List[str] , UpperCAmelCase : np.ndarray , UpperCAmelCase : Union[float, List[float]] , UpperCAmelCase : Union[float, List[float]] , UpperCAmelCase : Optional[Union[str, ChannelDimension]] = None , **UpperCAmelCase : Dict , ):
return normalize(UpperCAmelCase , mean=UpperCAmelCase , std=UpperCAmelCase , data_format=UpperCAmelCase , **UpperCAmelCase )
def __A ( self : Optional[int] , UpperCAmelCase : ImageInput , UpperCAmelCase : Optional[bool] = None , UpperCAmelCase : Dict[str, int] = None , UpperCAmelCase : PILImageResampling = None , UpperCAmelCase : bool = None , UpperCAmelCase : int = None , UpperCAmelCase : Optional[bool] = None , UpperCAmelCase : Optional[float] = None , UpperCAmelCase : Optional[bool] = None , UpperCAmelCase : Optional[Union[float, List[float]]] = None , UpperCAmelCase : Optional[Union[float, List[float]]] = None , UpperCAmelCase : Optional[Union[str, TensorType]] = None , UpperCAmelCase : Union[str, ChannelDimension] = ChannelDimension.FIRST , **UpperCAmelCase : str , ):
A_ = do_resize if do_resize is not None else self.do_resize
A_ = do_rescale if do_rescale is not None else self.do_rescale
A_ = do_normalize if do_normalize is not None else self.do_normalize
A_ = do_center_crop if do_center_crop is not None else self.do_center_crop
A_ = crop_size if crop_size is not None else self.crop_size
A_ = get_size_dict(UpperCAmelCase , param_name="crop_size" , default_to_square=UpperCAmelCase )
A_ = resample if resample is not None else self.resample
A_ = rescale_factor if rescale_factor is not None else self.rescale_factor
A_ = image_mean if image_mean is not None else self.image_mean
A_ = image_std if image_std is not None else self.image_std
A_ = size if size is not None else self.size
A_ = get_size_dict(UpperCAmelCase )
if not is_batched(UpperCAmelCase ):
A_ = [images]
if not valid_images(UpperCAmelCase ):
raise ValueError(
"Invalid image type. Must be of type PIL.Image.Image, numpy.ndarray, "
"torch.Tensor, tf.Tensor or jax.ndarray." )
if do_resize and size is None:
raise ValueError("Size must be specified if do_resize is True." )
if do_center_crop and crop_size is None:
raise ValueError("Crop size must be specified if do_center_crop is True." )
if do_rescale and rescale_factor is None:
raise ValueError("Rescale factor must be specified if do_rescale is True." )
# All transformations expect numpy arrays.
A_ = [to_numpy_array(UpperCAmelCase ) for image in images]
if do_resize:
A_ = [self.resize(image=UpperCAmelCase , size=UpperCAmelCase , resample=UpperCAmelCase ) for image in images]
if do_center_crop:
A_ = [self.center_crop(image=UpperCAmelCase , size=UpperCAmelCase ) for image in images]
if do_rescale:
A_ = [self.rescale(image=UpperCAmelCase , scale=UpperCAmelCase ) for image in images]
if do_normalize:
A_ = [self.normalize(image=UpperCAmelCase , mean=UpperCAmelCase , std=UpperCAmelCase ) for image in images]
A_ = [to_channel_dimension_format(UpperCAmelCase , UpperCAmelCase ) for image in images]
A_ = {"pixel_values": images}
return BatchFeature(data=UpperCAmelCase , tensor_type=UpperCAmelCase ) | 86 |
__a :Optional[int] = [sum(int(c, 10) ** 2 for c in i.__str__()) for i in range(10_0000)]
def __snake_case ( __UpperCamelCase : int ):
"""simple docstring"""
A_ = 0
while number:
# Increased Speed Slightly by checking every 5 digits together.
sum_of_digits_squared += DIGITS_SQUARED[number % 10_0000]
number //= 10_0000
return sum_of_digits_squared
# There are 2 Chains made,
# One ends with 89 with the chain member 58 being the one which when declared first,
# there will be the least number of iterations for all the members to be checked.
# The other one ends with 1 and has only one element 1.
# So 58 and 1 are chosen to be declared at the starting.
# Changed dictionary to an array to quicken the solution
__a :list[bool | None] = [None] * 1000_0000
__a :Optional[Any] = True
__a :List[Any] = False
def __snake_case ( __UpperCamelCase : int ):
"""simple docstring"""
if CHAINS[number - 1] is not None:
return CHAINS[number - 1] # type: ignore
A_ = chain(next_number(__UpperCamelCase ) )
A_ = number_chain
while number < 1000_0000:
A_ = number_chain
number *= 10
return number_chain
def __snake_case ( __UpperCamelCase : int = 1000_0000 ):
"""simple docstring"""
for i in range(1 ,__UpperCamelCase ):
if CHAINS[i] is None:
chain(i + 1 )
return CHAINS[:number].count(__UpperCamelCase )
if __name__ == "__main__":
import doctest
doctest.testmod()
print(F"{solution() = }") | 86 | 1 |
import PIL.Image
import PIL.ImageOps
from packaging import version
from PIL import Image
if version.parse(version.parse(PIL.__version__).base_version) >= version.parse('9.1.0'):
__a :Optional[Any] = {
'linear': PIL.Image.Resampling.BILINEAR,
'bilinear': PIL.Image.Resampling.BILINEAR,
'bicubic': PIL.Image.Resampling.BICUBIC,
'lanczos': PIL.Image.Resampling.LANCZOS,
'nearest': PIL.Image.Resampling.NEAREST,
}
else:
__a :List[Any] = {
'linear': PIL.Image.LINEAR,
'bilinear': PIL.Image.BILINEAR,
'bicubic': PIL.Image.BICUBIC,
'lanczos': PIL.Image.LANCZOS,
'nearest': PIL.Image.NEAREST,
}
def __snake_case ( __UpperCamelCase : Union[str, Any] ):
"""simple docstring"""
A_ = (images / 2 + 0.5).clamp(0 ,1 )
A_ = images.cpu().permute(0 ,2 ,3 ,1 ).float().numpy()
A_ = numpy_to_pil(__UpperCamelCase )
return images
def __snake_case ( __UpperCamelCase : int ):
"""simple docstring"""
if images.ndim == 3:
A_ = images[None, ...]
A_ = (images * 255).round().astype("uint8" )
if images.shape[-1] == 1:
# special case for grayscale (single channel) images
A_ = [Image.fromarray(image.squeeze() ,mode="L" ) for image in images]
else:
A_ = [Image.fromarray(__UpperCamelCase ) for image in images]
return pil_images | 86 |
from typing import TYPE_CHECKING
from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_tf_available, is_torch_available
__a :List[Any] = {
'configuration_tapas': ['TAPAS_PRETRAINED_CONFIG_ARCHIVE_MAP', 'TapasConfig'],
'tokenization_tapas': ['TapasTokenizer'],
}
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Any = [
'TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST',
'TapasForMaskedLM',
'TapasForQuestionAnswering',
'TapasForSequenceClassification',
'TapasModel',
'TapasPreTrainedModel',
'load_tf_weights_in_tapas',
]
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Dict = [
'TF_TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST',
'TFTapasForMaskedLM',
'TFTapasForQuestionAnswering',
'TFTapasForSequenceClassification',
'TFTapasModel',
'TFTapasPreTrainedModel',
]
if TYPE_CHECKING:
from .configuration_tapas import TAPAS_PRETRAINED_CONFIG_ARCHIVE_MAP, TapasConfig
from .tokenization_tapas import TapasTokenizer
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tapas import (
TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST,
TapasForMaskedLM,
TapasForQuestionAnswering,
TapasForSequenceClassification,
TapasModel,
TapasPreTrainedModel,
load_tf_weights_in_tapas,
)
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tf_tapas import (
TF_TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST,
TFTapasForMaskedLM,
TFTapasForQuestionAnswering,
TFTapasForSequenceClassification,
TFTapasModel,
TFTapasPreTrainedModel,
)
else:
import sys
__a :str = _LazyModule(__name__, globals()['__file__'], _import_structure, module_spec=__spec__) | 86 | 1 |
import pickle
import unittest
import torch
from accelerate import Accelerator
from accelerate.state import AcceleratorState
from accelerate.test_utils import require_cpu
@require_cpu
class _a ( unittest.TestCase ):
"""simple docstring"""
def __A ( self : List[str] ):
A_ = torch.nn.Linear(10 , 10 )
A_ = torch.optim.SGD(model.parameters() , 0.1 )
A_ = Accelerator()
A_ = accelerator.prepare(UpperCAmelCase )
try:
pickle.loads(pickle.dumps(UpperCAmelCase ) )
except Exception as e:
self.fail(f'''Accelerated optimizer pickling failed with {e}''' )
AcceleratorState._reset_state() | 86 |
# Lint as: python3
import sys
from collections.abc import Mapping
from typing import TYPE_CHECKING, Dict, Optional
import numpy as np
import pyarrow as pa
from .. import config
from ..utils.logging import get_logger
from ..utils.py_utils import map_nested
from .formatting import TensorFormatter
if TYPE_CHECKING:
import jax
import jaxlib
__a :List[Any] = get_logger()
__a :Optional[dict] = None
class _a ( TensorFormatter[Mapping, 'jax.Array', Mapping] ):
"""simple docstring"""
def __init__( self : str , UpperCAmelCase : int=None , UpperCAmelCase : List[str]=None , **UpperCAmelCase : List[Any] ):
super().__init__(features=UpperCAmelCase )
import jax
from jaxlib.xla_client import Device
if isinstance(UpperCAmelCase , UpperCAmelCase ):
raise ValueError(
f'''Expected {device} to be a `str` not {type(UpperCAmelCase )}, as `jaxlib.xla_extension.Device` '''
"is not serializable neither with `pickle` nor with `dill`. Instead you can surround "
"the device with `str()` to get its string identifier that will be internally mapped "
"to the actual `jaxlib.xla_extension.Device`." )
A_ = device if isinstance(UpperCAmelCase , UpperCAmelCase ) else str(jax.devices()[0] )
# using global variable since `jaxlib.xla_extension.Device` is not serializable neither
# with `pickle` nor with `dill`, so we need to use a global variable instead
global DEVICE_MAPPING
if DEVICE_MAPPING is None:
A_ = self._map_devices_to_str()
if self.device not in list(DEVICE_MAPPING.keys() ):
logger.warning(
f'''Device with string identifier {self.device} not listed among the available '''
f'''devices: {list(DEVICE_MAPPING.keys() )}, so falling back to the default '''
f'''device: {str(jax.devices()[0] )}.''' )
A_ = str(jax.devices()[0] )
A_ = jnp_array_kwargs
@staticmethod
def __A ( ):
import jax
return {str(UpperCAmelCase ): device for device in jax.devices()}
def __A ( self : Optional[int] , UpperCAmelCase : int ):
import jax
import jax.numpy as jnp
if isinstance(UpperCAmelCase , UpperCAmelCase ) and column:
if all(
isinstance(UpperCAmelCase , jax.Array ) and x.shape == column[0].shape and x.dtype == column[0].dtype for x in column ):
return jnp.stack(UpperCAmelCase , axis=0 )
return column
def __A ( self : List[str] , UpperCAmelCase : str ):
import jax
import jax.numpy as jnp
if isinstance(UpperCAmelCase , (str, bytes, type(UpperCAmelCase )) ):
return value
elif isinstance(UpperCAmelCase , (np.character, np.ndarray) ) and np.issubdtype(value.dtype , np.character ):
return value.tolist()
A_ = {}
if isinstance(UpperCAmelCase , (np.number, np.ndarray) ) and np.issubdtype(value.dtype , np.integer ):
# the default int precision depends on the jax config
# see https://jax.readthedocs.io/en/latest/notebooks/Common_Gotchas_in_JAX.html#double-64bit-precision
if jax.config.jax_enable_xaa:
A_ = {"dtype": jnp.intaa}
else:
A_ = {"dtype": jnp.intaa}
elif isinstance(UpperCAmelCase , (np.number, np.ndarray) ) and np.issubdtype(value.dtype , np.floating ):
A_ = {"dtype": jnp.floataa}
elif config.PIL_AVAILABLE and "PIL" in sys.modules:
import PIL.Image
if isinstance(UpperCAmelCase , PIL.Image.Image ):
A_ = np.asarray(UpperCAmelCase )
# using global variable since `jaxlib.xla_extension.Device` is not serializable neither
# with `pickle` nor with `dill`, so we need to use a global variable instead
global DEVICE_MAPPING
if DEVICE_MAPPING is None:
A_ = self._map_devices_to_str()
with jax.default_device(DEVICE_MAPPING[self.device] ):
# calling jnp.array on a np.ndarray does copy the data
# see https://github.com/google/jax/issues/4486
return jnp.array(UpperCAmelCase , **{**default_dtype, **self.jnp_array_kwargs} )
def __A ( self : Any , UpperCAmelCase : Dict ):
import jax
# support for torch, tf, jax etc.
if config.TORCH_AVAILABLE and "torch" in sys.modules:
import torch
if isinstance(UpperCAmelCase , torch.Tensor ):
return self._tensorize(data_struct.detach().cpu().numpy()[()] )
if hasattr(UpperCAmelCase , "__array__" ) and not isinstance(UpperCAmelCase , jax.Array ):
A_ = data_struct.__array__()
# support for nested types like struct of list of struct
if isinstance(UpperCAmelCase , np.ndarray ):
if data_struct.dtype == object: # jax arrays cannot be instantied from an array of objects
return self._consolidate([self.recursive_tensorize(UpperCAmelCase ) for substruct in data_struct] )
elif isinstance(UpperCAmelCase , (list, tuple) ):
return self._consolidate([self.recursive_tensorize(UpperCAmelCase ) for substruct in data_struct] )
return self._tensorize(UpperCAmelCase )
def __A ( self : Tuple , UpperCAmelCase : dict ):
return map_nested(self._recursive_tensorize , UpperCAmelCase , map_list=UpperCAmelCase )
def __A ( self : Dict , UpperCAmelCase : pa.Table ):
A_ = self.numpy_arrow_extractor().extract_row(UpperCAmelCase )
A_ = self.python_features_decoder.decode_row(UpperCAmelCase )
return self.recursive_tensorize(UpperCAmelCase )
def __A ( self : Any , UpperCAmelCase : pa.Table ):
A_ = self.numpy_arrow_extractor().extract_column(UpperCAmelCase )
A_ = self.python_features_decoder.decode_column(UpperCAmelCase , pa_table.column_names[0] )
A_ = self.recursive_tensorize(UpperCAmelCase )
A_ = self._consolidate(UpperCAmelCase )
return column
def __A ( self : Dict , UpperCAmelCase : pa.Table ):
A_ = self.numpy_arrow_extractor().extract_batch(UpperCAmelCase )
A_ = self.python_features_decoder.decode_batch(UpperCAmelCase )
A_ = self.recursive_tensorize(UpperCAmelCase )
for column_name in batch:
A_ = self._consolidate(batch[column_name] )
return batch | 86 | 1 |
import argparse
import torch
from transformers import (
SpeechTaConfig,
SpeechTaFeatureExtractor,
SpeechTaForSpeechToSpeech,
SpeechTaForSpeechToText,
SpeechTaForTextToSpeech,
SpeechTaProcessor,
SpeechTaTokenizer,
logging,
)
from transformers.tokenization_utils import AddedToken
logging.set_verbosity_info()
__a :str = logging.get_logger('transformers.models.speecht5')
__a :Optional[Any] = {
'speech_encoder_prenet.layer_norm': 'speecht5.encoder.prenet.feature_projection.layer_norm',
'speech_encoder_prenet.post_extract_proj': 'speecht5.encoder.prenet.feature_projection.projection',
'speech_encoder_prenet.pos_conv.0': 'speecht5.encoder.prenet.pos_conv_embed.conv',
'speech_encoder_prenet.mask_emb': 'speecht5.encoder.prenet.masked_spec_embed',
}
__a :Dict = {
'text_encoder_prenet.encoder_prenet.0': 'speecht5.encoder.prenet.embed_tokens',
'text_encoder_prenet.encoder_prenet.1.alpha': 'speecht5.encoder.prenet.encode_positions.alpha',
}
__a :Any = {
'speech_decoder_prenet.decoder_prenet.0.0.prenet.0.0': 'speecht5.decoder.prenet.layers.0',
'speech_decoder_prenet.decoder_prenet.0.0.prenet.1.0': 'speecht5.decoder.prenet.layers.1',
'speech_decoder_prenet.decoder_prenet.0.1': 'speecht5.decoder.prenet.final_layer',
'speech_decoder_prenet.decoder_prenet.1.alpha': 'speecht5.decoder.prenet.encode_positions.alpha',
'speech_decoder_prenet.spkembs_layer.0': 'speecht5.decoder.prenet.speaker_embeds_layer',
}
__a :int = {
'speech_decoder_postnet.feat_out': 'speech_decoder_postnet.feat_out',
'speech_decoder_postnet.prob_out': 'speech_decoder_postnet.prob_out',
'speech_decoder_postnet.postnet.postnet.0.0': 'speech_decoder_postnet.layers.0.conv',
'speech_decoder_postnet.postnet.postnet.0.1': 'speech_decoder_postnet.layers.0.batch_norm',
'speech_decoder_postnet.postnet.postnet.1.0': 'speech_decoder_postnet.layers.1.conv',
'speech_decoder_postnet.postnet.postnet.1.1': 'speech_decoder_postnet.layers.1.batch_norm',
'speech_decoder_postnet.postnet.postnet.2.0': 'speech_decoder_postnet.layers.2.conv',
'speech_decoder_postnet.postnet.postnet.2.1': 'speech_decoder_postnet.layers.2.batch_norm',
'speech_decoder_postnet.postnet.postnet.3.0': 'speech_decoder_postnet.layers.3.conv',
'speech_decoder_postnet.postnet.postnet.3.1': 'speech_decoder_postnet.layers.3.batch_norm',
'speech_decoder_postnet.postnet.postnet.4.0': 'speech_decoder_postnet.layers.4.conv',
'speech_decoder_postnet.postnet.postnet.4.1': 'speech_decoder_postnet.layers.4.batch_norm',
}
__a :Dict = {
'text_decoder_prenet.embed_tokens': 'speecht5.decoder.prenet.embed_tokens',
}
__a :Union[str, Any] = {
'text_decoder_postnet.output_projection': 'text_decoder_postnet.lm_head',
}
__a :int = {
'encoder.layers.*.self_attn.k_proj': 'speecht5.encoder.wrapped_encoder.layers.*.attention.k_proj',
'encoder.layers.*.self_attn.v_proj': 'speecht5.encoder.wrapped_encoder.layers.*.attention.v_proj',
'encoder.layers.*.self_attn.q_proj': 'speecht5.encoder.wrapped_encoder.layers.*.attention.q_proj',
'encoder.layers.*.self_attn.out_proj': 'speecht5.encoder.wrapped_encoder.layers.*.attention.out_proj',
'encoder.layers.*.self_attn_layer_norm': 'speecht5.encoder.wrapped_encoder.layers.*.layer_norm',
'encoder.layers.*.fc1': 'speecht5.encoder.wrapped_encoder.layers.*.feed_forward.intermediate_dense',
'encoder.layers.*.fc2': 'speecht5.encoder.wrapped_encoder.layers.*.feed_forward.output_dense',
'encoder.layers.*.final_layer_norm': 'speecht5.encoder.wrapped_encoder.layers.*.final_layer_norm',
'encoder.layer_norm': 'speecht5.encoder.wrapped_encoder.layer_norm',
'encoder.pos_emb.pe_k': 'speecht5.encoder.wrapped_encoder.embed_positions.pe_k',
}
__a :List[str] = {
'decoder.layers.*.self_attn.k_proj': 'speecht5.decoder.wrapped_decoder.layers.*.self_attn.k_proj',
'decoder.layers.*.self_attn.v_proj': 'speecht5.decoder.wrapped_decoder.layers.*.self_attn.v_proj',
'decoder.layers.*.self_attn.q_proj': 'speecht5.decoder.wrapped_decoder.layers.*.self_attn.q_proj',
'decoder.layers.*.self_attn.out_proj': 'speecht5.decoder.wrapped_decoder.layers.*.self_attn.out_proj',
'decoder.layers.*.self_attn_layer_norm': 'speecht5.decoder.wrapped_decoder.layers.*.self_attn_layer_norm',
'decoder.layers.*.encoder_attn.k_proj': 'speecht5.decoder.wrapped_decoder.layers.*.encoder_attn.k_proj',
'decoder.layers.*.encoder_attn.v_proj': 'speecht5.decoder.wrapped_decoder.layers.*.encoder_attn.v_proj',
'decoder.layers.*.encoder_attn.q_proj': 'speecht5.decoder.wrapped_decoder.layers.*.encoder_attn.q_proj',
'decoder.layers.*.encoder_attn.out_proj': 'speecht5.decoder.wrapped_decoder.layers.*.encoder_attn.out_proj',
'decoder.layers.*.encoder_attn_layer_norm': 'speecht5.decoder.wrapped_decoder.layers.*.encoder_attn_layer_norm',
'decoder.layers.*.fc1': 'speecht5.decoder.wrapped_decoder.layers.*.feed_forward.intermediate_dense',
'decoder.layers.*.fc2': 'speecht5.decoder.wrapped_decoder.layers.*.feed_forward.output_dense',
'decoder.layers.*.final_layer_norm': 'speecht5.decoder.wrapped_decoder.layers.*.final_layer_norm',
}
__a :List[str] = {
**MAPPING_SPEECH_ENCODER_PRENET,
**MAPPING_ENCODER,
**MAPPING_DECODER,
**MAPPING_TEXT_DECODER_PRENET,
**MAPPING_TEXT_DECODER_POSTNET,
}
__a :str = {
**MAPPING_TEXT_ENCODER_PRENET,
**MAPPING_ENCODER,
**MAPPING_DECODER,
**MAPPING_SPEECH_DECODER_PRENET,
**MAPPING_SPEECH_DECODER_POSTNET,
}
__a :Any = {
**MAPPING_SPEECH_ENCODER_PRENET,
**MAPPING_ENCODER,
**MAPPING_DECODER,
**MAPPING_SPEECH_DECODER_PRENET,
**MAPPING_SPEECH_DECODER_POSTNET,
}
__a :Optional[int] = []
__a :List[str] = [
'encoder.version',
'encoder.layers.*.norm_k.weight',
'encoder.layers.*.norm_k.bias',
'decoder.version',
'decoder.layers.*.norm_k.weight',
'decoder.layers.*.norm_k.bias',
'decoder.pos_emb.pe_k',
'speech_encoder_prenet.embed_positions._float_tensor',
'text_decoder_prenet.embed_positions._float_tensor',
]
__a :Union[str, Any] = IGNORE_KEYS + [
'encoder.proj',
'text_encoder_prenet.*',
'speech_decoder_prenet.*',
'speech_decoder_postnet.*',
]
__a :str = IGNORE_KEYS + [
'encoder.proj',
'speech_encoder_prenet.*',
'text_decoder_prenet.*',
'text_decoder_postnet.*',
]
__a :Dict = IGNORE_KEYS + [
'encoder.proj',
'text_encoder_prenet.*',
'text_decoder_prenet.*',
'text_decoder_postnet.*',
]
def __snake_case ( __UpperCamelCase : Optional[int] ,__UpperCamelCase : str ,__UpperCamelCase : Any ,__UpperCamelCase : Tuple ,__UpperCamelCase : Optional[int] ):
"""simple docstring"""
for attribute in key.split("." ):
A_ = getattr(__UpperCamelCase ,__UpperCamelCase )
if weight_type is not None:
A_ = getattr(__UpperCamelCase ,__UpperCamelCase ).shape
else:
A_ = hf_pointer.shape
if hf_shape != value.shape:
raise ValueError(
f'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
f''' {value.shape} for {full_name}''' )
if weight_type == "weight":
A_ = value
elif weight_type == "weight_g":
A_ = value
elif weight_type == "weight_v":
A_ = value
elif weight_type == "bias":
A_ = value
elif weight_type == "running_mean":
A_ = value
elif weight_type == "running_var":
A_ = value
elif weight_type == "num_batches_tracked":
A_ = value
else:
A_ = value
logger.info(f'''{key + ("." + weight_type if weight_type is not None else "")} was initialized from {full_name}.''' )
def __snake_case ( __UpperCamelCase : Tuple ,__UpperCamelCase : Optional[int] ):
"""simple docstring"""
for key in ignore_keys:
if key.endswith(".*" ):
if name.startswith(key[:-1] ):
return True
elif ".*." in key:
A_ , A_ = key.split(".*." )
if prefix in name and suffix in name:
return True
elif key in name:
return True
return False
def __snake_case ( __UpperCamelCase : Any ,__UpperCamelCase : List[str] ,__UpperCamelCase : Dict ):
"""simple docstring"""
A_ = []
if task == "s2t":
A_ = hf_model.speechta.encoder.prenet.feature_encoder
A_ = MAPPING_S2T
A_ = IGNORE_KEYS_S2T
elif task == "t2s":
A_ = None
A_ = MAPPING_T2S
A_ = IGNORE_KEYS_T2S
elif task == "s2s":
A_ = hf_model.speechta.encoder.prenet.feature_encoder
A_ = MAPPING_S2S
A_ = IGNORE_KEYS_S2S
else:
raise ValueError(f'''Unsupported task: {task}''' )
for name, value in fairseq_dict.items():
if should_ignore(__UpperCamelCase ,__UpperCamelCase ):
logger.info(f'''{name} was ignored''' )
continue
A_ = False
if "conv_layers" in name:
load_conv_layer(
__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,hf_model.config.feat_extract_norm == "group" ,)
A_ = True
else:
for key, mapped_key in MAPPING.items():
# mapped_key = "speecht5." + mapped_key if mapped_key not in TOP_LEVEL_KEYS else mapped_key
if "*" in key:
A_ , A_ = key.split(".*." )
if prefix in name and suffix in name:
A_ = suffix
# if key in name or key.split("w2v_model.")[-1] == name.split(".")[0]:
if key in name:
A_ = True
if "*" in mapped_key:
A_ = name.split(__UpperCamelCase )[0].split("." )[-2]
A_ = mapped_key.replace("*" ,__UpperCamelCase )
if "weight_g" in name:
A_ = "weight_g"
elif "weight_v" in name:
A_ = "weight_v"
elif "bias" in name:
A_ = "bias"
elif "weight" in name:
A_ = "weight"
elif "running_mean" in name:
A_ = "running_mean"
elif "running_var" in name:
A_ = "running_var"
elif "num_batches_tracked" in name:
A_ = "num_batches_tracked"
else:
A_ = None
set_recursively(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
continue
if not is_used:
unused_weights.append(__UpperCamelCase )
logger.warning(f'''Unused weights: {unused_weights}''' )
def __snake_case ( __UpperCamelCase : List[Any] ,__UpperCamelCase : Optional[Any] ,__UpperCamelCase : str ,__UpperCamelCase : List[str] ,__UpperCamelCase : Dict ):
"""simple docstring"""
A_ = full_name.split("conv_layers." )[-1]
A_ = name.split("." )
A_ = int(items[0] )
A_ = int(items[1] )
if type_id == 0:
if "bias" in name:
if value.shape != feature_extractor.conv_layers[layer_id].conv.bias.data.shape:
raise ValueError(
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.''' )
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
if value.shape != feature_extractor.conv_layers[layer_id].conv.weight.data.shape:
raise ValueError(
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.''' )
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
if value.shape != feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape:
raise ValueError(
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape} was found.''' )
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
if value.shape != feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape:
raise ValueError(
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape} was found.''' )
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(__UpperCamelCase )
@torch.no_grad()
def __snake_case ( __UpperCamelCase : Any ,__UpperCamelCase : str ,__UpperCamelCase : Tuple ,__UpperCamelCase : List[Any]=None ,__UpperCamelCase : List[Any]=None ,__UpperCamelCase : Tuple=None ,):
"""simple docstring"""
if config_path is not None:
A_ = SpeechTaConfig.from_pretrained(__UpperCamelCase )
else:
A_ = SpeechTaConfig()
if task == "s2t":
A_ = config.max_text_positions
A_ = SpeechTaForSpeechToText(__UpperCamelCase )
elif task == "t2s":
A_ = 1876
A_ = 600
A_ = config.max_speech_positions
A_ = SpeechTaForTextToSpeech(__UpperCamelCase )
elif task == "s2s":
A_ = 1876
A_ = config.max_speech_positions
A_ = SpeechTaForSpeechToSpeech(__UpperCamelCase )
else:
raise ValueError(f'''Unknown task name: {task}''' )
if vocab_path:
A_ = SpeechTaTokenizer(__UpperCamelCase ,model_max_length=config.max_text_positions )
# Mask token behaves like a normal word, i.e. include the space before it
A_ = AddedToken("<mask>" ,lstrip=__UpperCamelCase ,rstrip=__UpperCamelCase )
A_ = mask_token
tokenizer.add_special_tokens({"mask_token": mask_token} )
tokenizer.add_tokens(["<ctc_blank>"] )
A_ = SpeechTaFeatureExtractor()
A_ = SpeechTaProcessor(tokenizer=__UpperCamelCase ,feature_extractor=__UpperCamelCase )
processor.save_pretrained(__UpperCamelCase )
A_ = torch.load(__UpperCamelCase )
recursively_load_weights(fairseq_checkpoint["model"] ,__UpperCamelCase ,__UpperCamelCase )
model.save_pretrained(__UpperCamelCase )
if repo_id:
print("Pushing to the hub..." )
processor.push_to_hub(__UpperCamelCase )
model.push_to_hub(__UpperCamelCase )
if __name__ == "__main__":
__a :Union[str, Any] = argparse.ArgumentParser()
parser.add_argument(
'--task',
default='s2t',
type=str,
help='Type of the SpeechT5 model you\'d like to convert. Should be one of \'s2t\', \'t2s\', \'s2s\'.',
)
parser.add_argument('--checkpoint_path', required=True, default=None, type=str, help='Path to fairseq checkpoint')
parser.add_argument('--vocab_path', default=None, type=str, help='Path to SentencePiece model')
parser.add_argument('--config_path', default=None, type=str, help='Path to hf config.json of model to convert')
parser.add_argument(
'--pytorch_dump_folder_path', required=True, default=None, type=str, help='Path to the output PyTorch model.'
)
parser.add_argument(
'--push_to_hub', default=None, type=str, help='Where to upload the converted model on the 🤗 hub.'
)
__a :int = parser.parse_args()
convert_speechta_checkpoint(
args.task,
args.checkpoint_path,
args.pytorch_dump_folder_path,
args.config_path,
args.vocab_path,
args.push_to_hub,
) | 86 |
import logging
import os
from typing import List, Tuple
import numpy as np
import psutil
import torch
import torch.distributed as dist
from transformers import RagRetriever
__a :Any = logging.getLogger(__name__)
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Dict , UpperCAmelCase : List[Any] , UpperCAmelCase : Optional[int] , UpperCAmelCase : Optional[Any] , UpperCAmelCase : str=None ):
super().__init__(
UpperCAmelCase , question_encoder_tokenizer=UpperCAmelCase , generator_tokenizer=UpperCAmelCase , index=UpperCAmelCase , init_retrieval=UpperCAmelCase , )
A_ = None
def __A ( self : Dict , UpperCAmelCase : int ):
logger.info("initializing retrieval" )
# initializing a separate process group for retrieval as the default
# nccl backend doesn't support gather/scatter operations while gloo
# is too slow to replace nccl for the core gpu communication
if dist.is_initialized():
logger.info("dist initialized" )
# needs to be set manually
A_ = self._infer_socket_ifname()
# avoid clash with the NCCL port
A_ = str(distributed_port + 1 )
A_ = dist.new_group(ranks=UpperCAmelCase , backend="gloo" )
# initialize retriever only on the main worker
if not dist.is_initialized() or self._is_main():
logger.info("dist not initialized / main" )
self.index.init_index()
# all processes wait untill the retriever is initialized by the main process
if dist.is_initialized():
torch.distributed.barrier(group=self.process_group )
def __A ( self : List[str] ):
return dist.get_rank(group=self.process_group ) == 0
def __A ( self : Any , UpperCAmelCase : List[Any] , UpperCAmelCase : Union[str, Any] , UpperCAmelCase : Dict=torch.floataa ):
A_ = torch.empty(UpperCAmelCase , dtype=UpperCAmelCase )
dist.scatter(UpperCAmelCase , src=0 , scatter_list=UpperCAmelCase , group=self.process_group )
return target_tensor
def __A ( self : Any ):
A_ = psutil.net_if_addrs()
# a hacky way to deal with varying network interface names
A_ = next((addr for addr in addrs if addr.startswith("e" )) , UpperCAmelCase )
return ifname
def __A ( self : Tuple , UpperCAmelCase : np.ndarray , UpperCAmelCase : int ):
# single GPU training
if not dist.is_initialized():
A_ , A_ = self._main_retrieve(UpperCAmelCase , UpperCAmelCase )
return retrieved_doc_embeds, doc_ids, self.index.get_doc_dicts(UpperCAmelCase )
# distributed training
A_ = dist.get_world_size(group=self.process_group )
# gather logic
A_ = None
if self._is_main():
A_ = [torch.empty(question_hidden_states.shape , dtype=torch.floataa ) for _ in range(UpperCAmelCase )]
dist.gather(torch.tensor(UpperCAmelCase ) , dst=0 , gather_list=UpperCAmelCase , group=self.process_group )
# scatter logic
A_ = question_hidden_states.shape[0]
A_ = []
A_ = []
if self._is_main():
assert len(UpperCAmelCase ) == world_size
A_ , A_ = self._main_retrieve(torch.cat(UpperCAmelCase ).numpy() , UpperCAmelCase )
A_ , A_ = torch.tensor(UpperCAmelCase ), torch.tensor(UpperCAmelCase )
A_ = self._chunk_tensor(UpperCAmelCase , UpperCAmelCase )
A_ = self._chunk_tensor(UpperCAmelCase , UpperCAmelCase )
A_ = self._scattered(UpperCAmelCase , [n_queries, n_docs] , target_type=torch.intaa )
A_ = self._scattered(UpperCAmelCase , [n_queries, n_docs, question_hidden_states.shape[1]] )
return retrieved_doc_embeds.numpy(), doc_ids.numpy(), self.index.get_doc_dicts(UpperCAmelCase ) | 86 | 1 |
import inspect
import unittest
import warnings
from math import ceil, floor
from transformers import LevitConfig
from transformers.file_utils import cached_property, is_torch_available, is_vision_available
from transformers.models.auto import get_values
from transformers.testing_utils import require_torch, require_vision, slow, torch_device
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from transformers import (
MODEL_FOR_IMAGE_CLASSIFICATION_MAPPING,
MODEL_MAPPING,
LevitForImageClassification,
LevitForImageClassificationWithTeacher,
LevitModel,
)
from transformers.models.levit.modeling_levit import LEVIT_PRETRAINED_MODEL_ARCHIVE_LIST
if is_vision_available():
from PIL import Image
from transformers import LevitImageProcessor
class _a ( snake_case_ ):
"""simple docstring"""
def __A ( self : int ):
A_ = self.config_class(**self.inputs_dict )
self.parent.assertTrue(hasattr(UpperCAmelCase , "hidden_sizes" ) )
self.parent.assertTrue(hasattr(UpperCAmelCase , "num_attention_heads" ) )
class _a :
"""simple docstring"""
def __init__( self : Union[str, Any] , UpperCAmelCase : Optional[Any] , UpperCAmelCase : Any=13 , UpperCAmelCase : Any=64 , UpperCAmelCase : Any=3 , UpperCAmelCase : List[Any]=3 , UpperCAmelCase : int=2 , UpperCAmelCase : Optional[Any]=1 , UpperCAmelCase : Dict=16 , UpperCAmelCase : List[str]=[128, 256, 384] , UpperCAmelCase : List[Any]=[4, 6, 8] , UpperCAmelCase : Any=[2, 3, 4] , UpperCAmelCase : Dict=[16, 16, 16] , UpperCAmelCase : Optional[Any]=0 , UpperCAmelCase : str=[2, 2, 2] , UpperCAmelCase : int=[2, 2, 2] , UpperCAmelCase : Any=0.02 , UpperCAmelCase : Union[str, Any]=True , UpperCAmelCase : int=True , UpperCAmelCase : Tuple=2 , ):
A_ = parent
A_ = batch_size
A_ = image_size
A_ = num_channels
A_ = kernel_size
A_ = stride
A_ = padding
A_ = hidden_sizes
A_ = num_attention_heads
A_ = depths
A_ = key_dim
A_ = drop_path_rate
A_ = patch_size
A_ = attention_ratio
A_ = mlp_ratio
A_ = initializer_range
A_ = [
["Subsample", key_dim[0], hidden_sizes[0] // key_dim[0], 4, 2, 2],
["Subsample", key_dim[0], hidden_sizes[1] // key_dim[0], 4, 2, 2],
]
A_ = is_training
A_ = use_labels
A_ = num_labels
A_ = initializer_range
def __A ( self : List[str] ):
A_ = floats_tensor([self.batch_size, self.num_channels, self.image_size, self.image_size] )
A_ = None
if self.use_labels:
A_ = ids_tensor([self.batch_size] , self.num_labels )
A_ = self.get_config()
return config, pixel_values, labels
def __A ( self : Dict ):
return LevitConfig(
image_size=self.image_size , num_channels=self.num_channels , kernel_size=self.kernel_size , stride=self.stride , padding=self.padding , patch_size=self.patch_size , hidden_sizes=self.hidden_sizes , num_attention_heads=self.num_attention_heads , depths=self.depths , key_dim=self.key_dim , drop_path_rate=self.drop_path_rate , mlp_ratio=self.mlp_ratio , attention_ratio=self.attention_ratio , initializer_range=self.initializer_range , down_ops=self.down_ops , )
def __A ( self : Any , UpperCAmelCase : Tuple , UpperCAmelCase : Any , UpperCAmelCase : Tuple ):
A_ = LevitModel(config=UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
A_ = model(UpperCAmelCase )
A_ = (self.image_size, self.image_size)
A_ , A_ = image_size[0], image_size[1]
for _ in range(4 ):
A_ = floor(((height + 2 * self.padding - self.kernel_size) / self.stride) + 1 )
A_ = floor(((width + 2 * self.padding - self.kernel_size) / self.stride) + 1 )
self.parent.assertEqual(
result.last_hidden_state.shape , (self.batch_size, ceil(height / 4 ) * ceil(width / 4 ), self.hidden_sizes[-1]) , )
def __A ( self : str , UpperCAmelCase : Optional[int] , UpperCAmelCase : List[str] , UpperCAmelCase : List[str] ):
A_ = self.num_labels
A_ = LevitForImageClassification(UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
A_ = model(UpperCAmelCase , labels=UpperCAmelCase )
self.parent.assertEqual(result.logits.shape , (self.batch_size, self.num_labels) )
def __A ( self : str ):
A_ = self.prepare_config_and_inputs()
A_ , A_ , A_ = config_and_inputs
A_ = {"pixel_values": pixel_values}
return config, inputs_dict
@require_torch
class _a ( snake_case_ , snake_case_ , unittest.TestCase ):
"""simple docstring"""
_lowerCamelCase : Dict = (
(LevitModel, LevitForImageClassification, LevitForImageClassificationWithTeacher)
if is_torch_available()
else ()
)
_lowerCamelCase : int = (
{
'feature-extraction': LevitModel,
'image-classification': (LevitForImageClassification, LevitForImageClassificationWithTeacher),
}
if is_torch_available()
else {}
)
_lowerCamelCase : int = False
_lowerCamelCase : Any = False
_lowerCamelCase : Any = False
_lowerCamelCase : Dict = False
_lowerCamelCase : Dict = False
def __A ( self : int ):
A_ = LevitModelTester(self )
A_ = ConfigTester(self , config_class=UpperCAmelCase , has_text_modality=UpperCAmelCase , hidden_size=37 )
def __A ( self : Optional[int] ):
self.create_and_test_config_common_properties()
self.config_tester.create_and_test_config_to_json_string()
self.config_tester.create_and_test_config_to_json_file()
self.config_tester.create_and_test_config_from_and_save_pretrained()
self.config_tester.create_and_test_config_with_num_labels()
self.config_tester.check_config_can_be_init_without_params()
self.config_tester.check_config_arguments_init()
def __A ( self : List[str] ):
return
@unittest.skip(reason="Levit does not use inputs_embeds" )
def __A ( self : Dict ):
pass
@unittest.skip(reason="Levit does not support input and output embeddings" )
def __A ( self : str ):
pass
@unittest.skip(reason="Levit does not output attentions" )
def __A ( self : Dict ):
pass
def __A ( self : str ):
A_ , A_ = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
A_ = model_class(UpperCAmelCase )
A_ = inspect.signature(model.forward )
# signature.parameters is an OrderedDict => so arg_names order is deterministic
A_ = [*signature.parameters.keys()]
A_ = ["pixel_values"]
self.assertListEqual(arg_names[:1] , UpperCAmelCase )
def __A ( self : Dict ):
def check_hidden_states_output(UpperCAmelCase : Union[str, Any] , UpperCAmelCase : Union[str, Any] , UpperCAmelCase : Tuple ):
A_ = model_class(UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
with torch.no_grad():
A_ = model(**self._prepare_for_class(UpperCAmelCase , UpperCAmelCase ) )
A_ = outputs.hidden_states
A_ = len(self.model_tester.depths ) + 1
self.assertEqual(len(UpperCAmelCase ) , UpperCAmelCase )
A_ = (self.model_tester.image_size, self.model_tester.image_size)
A_ , A_ = image_size[0], image_size[1]
for _ in range(4 ):
A_ = floor(
(
(height + 2 * self.model_tester.padding - self.model_tester.kernel_size)
/ self.model_tester.stride
)
+ 1 )
A_ = floor(
(
(width + 2 * self.model_tester.padding - self.model_tester.kernel_size)
/ self.model_tester.stride
)
+ 1 )
# verify the first hidden states (first block)
self.assertListEqual(
list(hidden_states[0].shape[-2:] ) , [
height * width,
self.model_tester.hidden_sizes[0],
] , )
A_ , A_ = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
A_ = True
check_hidden_states_output(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase )
# check that output_hidden_states also work using config
del inputs_dict["output_hidden_states"]
A_ = True
check_hidden_states_output(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase )
@unittest.skip("Will be fixed soon by reducing the size of the model used for common tests." )
def __A ( self : Tuple ):
pass
def __A ( self : Any , UpperCAmelCase : Tuple , UpperCAmelCase : List[Any] , UpperCAmelCase : Dict=False ):
A_ = super()._prepare_for_class(UpperCAmelCase , UpperCAmelCase , return_labels=UpperCAmelCase )
if return_labels:
if model_class.__name__ == "LevitForImageClassificationWithTeacher":
del inputs_dict["labels"]
return inputs_dict
def __A ( self : int ):
A_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*UpperCAmelCase )
def __A ( self : Dict ):
A_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_image_classification(*UpperCAmelCase )
def __A ( self : Tuple ):
if not self.model_tester.is_training:
return
A_ , A_ = self.model_tester.prepare_config_and_inputs_for_common()
A_ = True
for model_class in self.all_model_classes:
# LevitForImageClassificationWithTeacher supports inference-only
if (
model_class in get_values(UpperCAmelCase )
or model_class.__name__ == "LevitForImageClassificationWithTeacher"
):
continue
A_ = model_class(UpperCAmelCase )
model.to(UpperCAmelCase )
model.train()
A_ = self._prepare_for_class(UpperCAmelCase , UpperCAmelCase , return_labels=UpperCAmelCase )
A_ = model(**UpperCAmelCase ).loss
loss.backward()
def __A ( self : Union[str, Any] ):
A_ , A_ = self.model_tester.prepare_config_and_inputs_for_common()
if not self.model_tester.is_training:
return
A_ = False
A_ = True
for model_class in self.all_model_classes:
if model_class in get_values(UpperCAmelCase ) or not model_class.supports_gradient_checkpointing:
continue
# LevitForImageClassificationWithTeacher supports inference-only
if model_class.__name__ == "LevitForImageClassificationWithTeacher":
continue
A_ = model_class(UpperCAmelCase )
model.gradient_checkpointing_enable()
model.to(UpperCAmelCase )
model.train()
A_ = self._prepare_for_class(UpperCAmelCase , UpperCAmelCase , return_labels=UpperCAmelCase )
A_ = model(**UpperCAmelCase ).loss
loss.backward()
def __A ( self : Dict ):
A_ , A_ = self.model_tester.prepare_config_and_inputs_for_common()
A_ = [
{"title": "multi_label_classification", "num_labels": 2, "dtype": torch.float},
{"title": "single_label_classification", "num_labels": 1, "dtype": torch.long},
{"title": "regression", "num_labels": 1, "dtype": torch.float},
]
for model_class in self.all_model_classes:
if (
model_class
not in [
*get_values(UpperCAmelCase ),
]
or model_class.__name__ == "LevitForImageClassificationWithTeacher"
):
continue
for problem_type in problem_types:
with self.subTest(msg=f'''Testing {model_class} with {problem_type["title"]}''' ):
A_ = problem_type["title"]
A_ = problem_type["num_labels"]
A_ = model_class(UpperCAmelCase )
model.to(UpperCAmelCase )
model.train()
A_ = self._prepare_for_class(UpperCAmelCase , UpperCAmelCase , return_labels=UpperCAmelCase )
if problem_type["num_labels"] > 1:
A_ = inputs["labels"].unsqueeze(1 ).repeat(1 , problem_type["num_labels"] )
A_ = inputs["labels"].to(problem_type["dtype"] )
# This tests that we do not trigger the warning form PyTorch "Using a target size that is different
# to the input size. This will likely lead to incorrect results due to broadcasting. Please ensure
# they have the same size." which is a symptom something in wrong for the regression problem.
# See https://github.com/huggingface/transformers/issues/11780
with warnings.catch_warnings(record=UpperCAmelCase ) as warning_list:
A_ = model(**UpperCAmelCase ).loss
for w in warning_list:
if "Using a target size that is different to the input size" in str(w.message ):
raise ValueError(
f'''Something is going wrong in the regression problem: intercepted {w.message}''' )
loss.backward()
@slow
def __A ( self : Optional[Any] ):
for model_name in LEVIT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
A_ = LevitModel.from_pretrained(UpperCAmelCase )
self.assertIsNotNone(UpperCAmelCase )
def __snake_case ( ):
"""simple docstring"""
A_ = Image.open("./tests/fixtures/tests_samples/COCO/000000039769.png" )
return image
@require_torch
@require_vision
class _a ( unittest.TestCase ):
"""simple docstring"""
@cached_property
def __A ( self : Any ):
return LevitImageProcessor.from_pretrained(LEVIT_PRETRAINED_MODEL_ARCHIVE_LIST[0] )
@slow
def __A ( self : Any ):
A_ = LevitForImageClassificationWithTeacher.from_pretrained(LEVIT_PRETRAINED_MODEL_ARCHIVE_LIST[0] ).to(
UpperCAmelCase )
A_ = self.default_image_processor
A_ = prepare_img()
A_ = image_processor(images=UpperCAmelCase , return_tensors="pt" ).to(UpperCAmelCase )
# forward pass
with torch.no_grad():
A_ = model(**UpperCAmelCase )
# verify the logits
A_ = torch.Size((1, 1000) )
self.assertEqual(outputs.logits.shape , UpperCAmelCase )
A_ = torch.tensor([1.0_448, -0.3_745, -1.8_317] ).to(UpperCAmelCase )
self.assertTrue(torch.allclose(outputs.logits[0, :3] , UpperCAmelCase , atol=1E-4 ) ) | 86 |
from jiwer import compute_measures
import datasets
__a :List[Any] = '\\n@inproceedings{inproceedings,\n author = {Morris, Andrew and Maier, Viktoria and Green, Phil},\n year = {2004},\n month = {01},\n pages = {},\n title = {From WER and RIL to MER and WIL: improved evaluation measures for connected speech recognition.}\n}\n'
__a :Union[str, Any] = '\\nWord error rate (WER) is a common metric of the performance of an automatic speech recognition system.\n\nThe general difficulty of measuring performance lies in the fact that the recognized word sequence can have a different length from the reference word sequence (supposedly the correct one). The WER is derived from the Levenshtein distance, working at the word level instead of the phoneme level. The WER is a valuable tool for comparing different systems as well as for evaluating improvements within one system. This kind of measurement, however, provides no details on the nature of translation errors and further work is therefore required to identify the main source(s) of error and to focus any research effort.\n\nThis problem is solved by first aligning the recognized word sequence with the reference (spoken) word sequence using dynamic string alignment. Examination of this issue is seen through a theory called the power law that states the correlation between perplexity and word error rate.\n\nWord error rate can then be computed as:\n\nWER = (S + D + I) / N = (S + D + I) / (S + D + C)\n\nwhere\n\nS is the number of substitutions,\nD is the number of deletions,\nI is the number of insertions,\nC is the number of correct words,\nN is the number of words in the reference (N=S+D+C).\n\nThis value indicates the average number of errors per reference word. The lower the value, the better the\nperformance of the ASR system with a WER of 0 being a perfect score.\n'
__a :str = '\nCompute WER score of transcribed segments against references.\n\nArgs:\n references: List of references for each speech input.\n predictions: List of transcriptions to score.\n concatenate_texts (bool, default=False): Whether to concatenate all input texts or compute WER iteratively.\n\nReturns:\n (float): the word error rate\n\nExamples:\n\n >>> predictions = ["this is the prediction", "there is an other sample"]\n >>> references = ["this is the reference", "there is another one"]\n >>> wer = datasets.load_metric("wer")\n >>> wer_score = wer.compute(predictions=predictions, references=references)\n >>> print(wer_score)\n 0.5\n'
@datasets.utils.file_utils.add_start_docstrings(_DESCRIPTION , _KWARGS_DESCRIPTION )
class _a ( datasets.Metric ):
"""simple docstring"""
def __A ( self : Any ):
return datasets.MetricInfo(
description=_DESCRIPTION , citation=_CITATION , inputs_description=_KWARGS_DESCRIPTION , features=datasets.Features(
{
"predictions": datasets.Value("string" , id="sequence" ),
"references": datasets.Value("string" , id="sequence" ),
} ) , codebase_urls=["https://github.com/jitsi/jiwer/"] , reference_urls=[
"https://en.wikipedia.org/wiki/Word_error_rate",
] , )
def __A ( self : Dict , UpperCAmelCase : Dict=None , UpperCAmelCase : List[Any]=None , UpperCAmelCase : str=False ):
if concatenate_texts:
return compute_measures(UpperCAmelCase , UpperCAmelCase )["wer"]
else:
A_ = 0
A_ = 0
for prediction, reference in zip(UpperCAmelCase , UpperCAmelCase ):
A_ = compute_measures(UpperCAmelCase , UpperCAmelCase )
incorrect += measures["substitutions"] + measures["deletions"] + measures["insertions"]
total += measures["substitutions"] + measures["deletions"] + measures["hits"]
return incorrect / total | 86 | 1 |
import warnings
from ..trainer import Trainer
from ..utils import logging
__a :Union[str, Any] = logging.get_logger(__name__)
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : int , UpperCAmelCase : Optional[Any]=None , **UpperCAmelCase : Union[str, Any] ):
warnings.warn(
"`SageMakerTrainer` is deprecated and will be removed in v5 of Transformers. You can use `Trainer` "
"instead." , UpperCAmelCase , )
super().__init__(args=UpperCAmelCase , **UpperCAmelCase ) | 86 |
class _a :
"""simple docstring"""
def __init__( self : Union[str, Any] , UpperCAmelCase : int , UpperCAmelCase : Any , UpperCAmelCase : Dict ):
A_ = None
A_ = None
A_ = graph
self._normalize_graph(UpperCAmelCase , UpperCAmelCase )
A_ = len(UpperCAmelCase )
A_ = None
def __A ( self : Union[str, Any] , UpperCAmelCase : List[str] , UpperCAmelCase : Tuple ):
if sources is int:
A_ = [sources]
if sinks is int:
A_ = [sinks]
if len(UpperCAmelCase ) == 0 or len(UpperCAmelCase ) == 0:
return
A_ = sources[0]
A_ = sinks[0]
# make fake vertex if there are more
# than one source or sink
if len(UpperCAmelCase ) > 1 or len(UpperCAmelCase ) > 1:
A_ = 0
for i in sources:
max_input_flow += sum(self.graph[i] )
A_ = len(self.graph ) + 1
for room in self.graph:
room.insert(0 , 0 )
self.graph.insert(0 , [0] * size )
for i in sources:
A_ = max_input_flow
A_ = 0
A_ = len(self.graph ) + 1
for room in self.graph:
room.append(0 )
self.graph.append([0] * size )
for i in sinks:
A_ = max_input_flow
A_ = size - 1
def __A ( self : str ):
if self.maximum_flow_algorithm is None:
raise Exception("You need to set maximum flow algorithm before." )
if self.source_index is None or self.sink_index is None:
return 0
self.maximum_flow_algorithm.execute()
return self.maximum_flow_algorithm.getMaximumFlow()
def __A ( self : Tuple , UpperCAmelCase : List[Any] ):
A_ = algorithm(self )
class _a :
"""simple docstring"""
def __init__( self : Tuple , UpperCAmelCase : List[str] ):
A_ = flow_network
A_ = flow_network.verticesCount
A_ = flow_network.sourceIndex
A_ = flow_network.sinkIndex
# it's just a reference, so you shouldn't change
# it in your algorithms, use deep copy before doing that
A_ = flow_network.graph
A_ = False
def __A ( self : Optional[int] ):
if not self.executed:
self._algorithm()
A_ = True
def __A ( self : Dict ):
pass
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Optional[Any] , UpperCAmelCase : List[Any] ):
super().__init__(UpperCAmelCase )
# use this to save your result
A_ = -1
def __A ( self : Tuple ):
if not self.executed:
raise Exception("You should execute algorithm before using its result!" )
return self.maximum_flow
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Tuple , UpperCAmelCase : Union[str, Any] ):
super().__init__(UpperCAmelCase )
A_ = [[0] * self.verticies_count for i in range(self.verticies_count )]
A_ = [0] * self.verticies_count
A_ = [0] * self.verticies_count
def __A ( self : List[str] ):
A_ = self.verticies_count
# push some substance to graph
for nextvertex_index, bandwidth in enumerate(self.graph[self.source_index] ):
self.preflow[self.source_index][nextvertex_index] += bandwidth
self.preflow[nextvertex_index][self.source_index] -= bandwidth
self.excesses[nextvertex_index] += bandwidth
# Relabel-to-front selection rule
A_ = [
i
for i in range(self.verticies_count )
if i != self.source_index and i != self.sink_index
]
# move through list
A_ = 0
while i < len(UpperCAmelCase ):
A_ = vertices_list[i]
A_ = self.heights[vertex_index]
self.process_vertex(UpperCAmelCase )
if self.heights[vertex_index] > previous_height:
# if it was relabeled, swap elements
# and start from 0 index
vertices_list.insert(0 , vertices_list.pop(UpperCAmelCase ) )
A_ = 0
else:
i += 1
A_ = sum(self.preflow[self.source_index] )
def __A ( self : List[str] , UpperCAmelCase : Dict ):
while self.excesses[vertex_index] > 0:
for neighbour_index in range(self.verticies_count ):
# if it's neighbour and current vertex is higher
if (
self.graph[vertex_index][neighbour_index]
- self.preflow[vertex_index][neighbour_index]
> 0
and self.heights[vertex_index] > self.heights[neighbour_index]
):
self.push(UpperCAmelCase , UpperCAmelCase )
self.relabel(UpperCAmelCase )
def __A ( self : Optional[Any] , UpperCAmelCase : Dict , UpperCAmelCase : str ):
A_ = min(
self.excesses[from_index] , self.graph[from_index][to_index] - self.preflow[from_index][to_index] , )
self.preflow[from_index][to_index] += preflow_delta
self.preflow[to_index][from_index] -= preflow_delta
self.excesses[from_index] -= preflow_delta
self.excesses[to_index] += preflow_delta
def __A ( self : Optional[Any] , UpperCAmelCase : List[Any] ):
A_ = None
for to_index in range(self.verticies_count ):
if (
self.graph[vertex_index][to_index]
- self.preflow[vertex_index][to_index]
> 0
) and (min_height is None or self.heights[to_index] < min_height):
A_ = self.heights[to_index]
if min_height is not None:
A_ = min_height + 1
if __name__ == "__main__":
__a :Tuple = [0]
__a :Tuple = [3]
# graph = [
# [0, 0, 4, 6, 0, 0],
# [0, 0, 5, 2, 0, 0],
# [0, 0, 0, 0, 4, 4],
# [0, 0, 0, 0, 6, 6],
# [0, 0, 0, 0, 0, 0],
# [0, 0, 0, 0, 0, 0],
# ]
__a :List[str] = [[0, 7, 0, 0], [0, 0, 6, 0], [0, 0, 0, 8], [9, 0, 0, 0]]
# prepare our network
__a :List[str] = FlowNetwork(graph, entrances, exits)
# set algorithm
flow_network.set_maximum_flow_algorithm(PushRelabelExecutor)
# and calculate
__a :List[Any] = flow_network.find_maximum_flow()
print(F"maximum flow is {maximum_flow}") | 86 | 1 |
from collections import deque
def __snake_case ( __UpperCamelCase : int ):
"""simple docstring"""
A_ = len(__UpperCamelCase )
A_ = deque()
A_ = [False for _ in range(__UpperCamelCase )]
A_ = [-1 for _ in range(__UpperCamelCase )]
A_ = index_of[:]
def strong_connect(__UpperCamelCase : List[str] ,__UpperCamelCase : str ,__UpperCamelCase : Union[str, Any] ):
A_ = index # the number when this node is seen
A_ = index # lowest rank node reachable from here
index += 1
stack.append(__UpperCamelCase )
A_ = True
for w in g[v]:
if index_of[w] == -1:
A_ = strong_connect(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
A_ = (
lowlink_of[w] if lowlink_of[w] < lowlink_of[v] else lowlink_of[v]
)
elif on_stack[w]:
A_ = (
lowlink_of[w] if lowlink_of[w] < lowlink_of[v] else lowlink_of[v]
)
if lowlink_of[v] == index_of[v]:
A_ = []
A_ = stack.pop()
A_ = False
component.append(__UpperCamelCase )
while w != v:
A_ = stack.pop()
A_ = False
component.append(__UpperCamelCase )
components.append(__UpperCamelCase )
return index
A_ = []
for v in range(__UpperCamelCase ):
if index_of[v] == -1:
strong_connect(__UpperCamelCase ,0 ,__UpperCamelCase )
return components
def __snake_case ( __UpperCamelCase : Any ,__UpperCamelCase : Tuple ):
"""simple docstring"""
A_ = [[] for _ in range(__UpperCamelCase )]
for u, v in edges:
g[u].append(__UpperCamelCase )
return g
if __name__ == "__main__":
# Test
__a :Optional[int] = 7
__a :Optional[int] = [0, 0, 1, 2, 3, 3, 4, 4, 6]
__a :str = [1, 3, 2, 0, 1, 4, 5, 6, 5]
__a :str = [(u, v) for u, v in zip(source, target)]
__a :int = create_graph(n_vertices, edges)
assert [[5], [6], [4], [3, 2, 1, 0]] == tarjan(g) | 86 |
from typing import TYPE_CHECKING
from ...utils import (
OptionalDependencyNotAvailable,
_LazyModule,
is_flax_available,
is_sentencepiece_available,
is_tf_available,
is_tokenizers_available,
is_torch_available,
)
__a :Dict = {'configuration_xglm': ['XGLM_PRETRAINED_CONFIG_ARCHIVE_MAP', 'XGLMConfig']}
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Dict = ['XGLMTokenizer']
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :str = ['XGLMTokenizerFast']
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Tuple = [
'XGLM_PRETRAINED_MODEL_ARCHIVE_LIST',
'XGLMForCausalLM',
'XGLMModel',
'XGLMPreTrainedModel',
]
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :List[Any] = [
'FlaxXGLMForCausalLM',
'FlaxXGLMModel',
'FlaxXGLMPreTrainedModel',
]
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Any = [
'TF_XGLM_PRETRAINED_MODEL_ARCHIVE_LIST',
'TFXGLMForCausalLM',
'TFXGLMModel',
'TFXGLMPreTrainedModel',
]
if TYPE_CHECKING:
from .configuration_xglm import XGLM_PRETRAINED_CONFIG_ARCHIVE_MAP, XGLMConfig
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_xglm import XGLMTokenizer
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_xglm_fast import XGLMTokenizerFast
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_xglm import XGLM_PRETRAINED_MODEL_ARCHIVE_LIST, XGLMForCausalLM, XGLMModel, XGLMPreTrainedModel
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_flax_xglm import FlaxXGLMForCausalLM, FlaxXGLMModel, FlaxXGLMPreTrainedModel
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tf_xglm import (
TF_XGLM_PRETRAINED_MODEL_ARCHIVE_LIST,
TFXGLMForCausalLM,
TFXGLMModel,
TFXGLMPreTrainedModel,
)
else:
import sys
__a :List[Any] = _LazyModule(__name__, globals()['__file__'], _import_structure) | 86 | 1 |
# Lint as: python3
import sys
from collections.abc import Mapping
from typing import TYPE_CHECKING, Dict, Optional
import numpy as np
import pyarrow as pa
from .. import config
from ..utils.logging import get_logger
from ..utils.py_utils import map_nested
from .formatting import TensorFormatter
if TYPE_CHECKING:
import jax
import jaxlib
__a :List[Any] = get_logger()
__a :Optional[dict] = None
class _a ( TensorFormatter[Mapping, 'jax.Array', Mapping] ):
"""simple docstring"""
def __init__( self : str , UpperCAmelCase : int=None , UpperCAmelCase : List[str]=None , **UpperCAmelCase : List[Any] ):
super().__init__(features=UpperCAmelCase )
import jax
from jaxlib.xla_client import Device
if isinstance(UpperCAmelCase , UpperCAmelCase ):
raise ValueError(
f'''Expected {device} to be a `str` not {type(UpperCAmelCase )}, as `jaxlib.xla_extension.Device` '''
"is not serializable neither with `pickle` nor with `dill`. Instead you can surround "
"the device with `str()` to get its string identifier that will be internally mapped "
"to the actual `jaxlib.xla_extension.Device`." )
A_ = device if isinstance(UpperCAmelCase , UpperCAmelCase ) else str(jax.devices()[0] )
# using global variable since `jaxlib.xla_extension.Device` is not serializable neither
# with `pickle` nor with `dill`, so we need to use a global variable instead
global DEVICE_MAPPING
if DEVICE_MAPPING is None:
A_ = self._map_devices_to_str()
if self.device not in list(DEVICE_MAPPING.keys() ):
logger.warning(
f'''Device with string identifier {self.device} not listed among the available '''
f'''devices: {list(DEVICE_MAPPING.keys() )}, so falling back to the default '''
f'''device: {str(jax.devices()[0] )}.''' )
A_ = str(jax.devices()[0] )
A_ = jnp_array_kwargs
@staticmethod
def __A ( ):
import jax
return {str(UpperCAmelCase ): device for device in jax.devices()}
def __A ( self : Optional[int] , UpperCAmelCase : int ):
import jax
import jax.numpy as jnp
if isinstance(UpperCAmelCase , UpperCAmelCase ) and column:
if all(
isinstance(UpperCAmelCase , jax.Array ) and x.shape == column[0].shape and x.dtype == column[0].dtype for x in column ):
return jnp.stack(UpperCAmelCase , axis=0 )
return column
def __A ( self : List[str] , UpperCAmelCase : str ):
import jax
import jax.numpy as jnp
if isinstance(UpperCAmelCase , (str, bytes, type(UpperCAmelCase )) ):
return value
elif isinstance(UpperCAmelCase , (np.character, np.ndarray) ) and np.issubdtype(value.dtype , np.character ):
return value.tolist()
A_ = {}
if isinstance(UpperCAmelCase , (np.number, np.ndarray) ) and np.issubdtype(value.dtype , np.integer ):
# the default int precision depends on the jax config
# see https://jax.readthedocs.io/en/latest/notebooks/Common_Gotchas_in_JAX.html#double-64bit-precision
if jax.config.jax_enable_xaa:
A_ = {"dtype": jnp.intaa}
else:
A_ = {"dtype": jnp.intaa}
elif isinstance(UpperCAmelCase , (np.number, np.ndarray) ) and np.issubdtype(value.dtype , np.floating ):
A_ = {"dtype": jnp.floataa}
elif config.PIL_AVAILABLE and "PIL" in sys.modules:
import PIL.Image
if isinstance(UpperCAmelCase , PIL.Image.Image ):
A_ = np.asarray(UpperCAmelCase )
# using global variable since `jaxlib.xla_extension.Device` is not serializable neither
# with `pickle` nor with `dill`, so we need to use a global variable instead
global DEVICE_MAPPING
if DEVICE_MAPPING is None:
A_ = self._map_devices_to_str()
with jax.default_device(DEVICE_MAPPING[self.device] ):
# calling jnp.array on a np.ndarray does copy the data
# see https://github.com/google/jax/issues/4486
return jnp.array(UpperCAmelCase , **{**default_dtype, **self.jnp_array_kwargs} )
def __A ( self : Any , UpperCAmelCase : Dict ):
import jax
# support for torch, tf, jax etc.
if config.TORCH_AVAILABLE and "torch" in sys.modules:
import torch
if isinstance(UpperCAmelCase , torch.Tensor ):
return self._tensorize(data_struct.detach().cpu().numpy()[()] )
if hasattr(UpperCAmelCase , "__array__" ) and not isinstance(UpperCAmelCase , jax.Array ):
A_ = data_struct.__array__()
# support for nested types like struct of list of struct
if isinstance(UpperCAmelCase , np.ndarray ):
if data_struct.dtype == object: # jax arrays cannot be instantied from an array of objects
return self._consolidate([self.recursive_tensorize(UpperCAmelCase ) for substruct in data_struct] )
elif isinstance(UpperCAmelCase , (list, tuple) ):
return self._consolidate([self.recursive_tensorize(UpperCAmelCase ) for substruct in data_struct] )
return self._tensorize(UpperCAmelCase )
def __A ( self : Tuple , UpperCAmelCase : dict ):
return map_nested(self._recursive_tensorize , UpperCAmelCase , map_list=UpperCAmelCase )
def __A ( self : Dict , UpperCAmelCase : pa.Table ):
A_ = self.numpy_arrow_extractor().extract_row(UpperCAmelCase )
A_ = self.python_features_decoder.decode_row(UpperCAmelCase )
return self.recursive_tensorize(UpperCAmelCase )
def __A ( self : Any , UpperCAmelCase : pa.Table ):
A_ = self.numpy_arrow_extractor().extract_column(UpperCAmelCase )
A_ = self.python_features_decoder.decode_column(UpperCAmelCase , pa_table.column_names[0] )
A_ = self.recursive_tensorize(UpperCAmelCase )
A_ = self._consolidate(UpperCAmelCase )
return column
def __A ( self : Dict , UpperCAmelCase : pa.Table ):
A_ = self.numpy_arrow_extractor().extract_batch(UpperCAmelCase )
A_ = self.python_features_decoder.decode_batch(UpperCAmelCase )
A_ = self.recursive_tensorize(UpperCAmelCase )
for column_name in batch:
A_ = self._consolidate(batch[column_name] )
return batch | 86 |
# Usage:
# ./gen-card-facebook-wmt19.py
import os
from pathlib import Path
def __snake_case ( __UpperCamelCase : str ,__UpperCamelCase : str ,__UpperCamelCase : List[str] ):
"""simple docstring"""
A_ = {
"en": "Machine learning is great, isn't it?",
"ru": "Машинное обучение - это здорово, не так ли?",
"de": "Maschinelles Lernen ist großartig, oder?",
}
# BLUE scores as follows:
# "pair": [fairseq, transformers]
A_ = {
"ru-en": ["[41.3](http://matrix.statmt.org/matrix/output/1907?run_id=6937)", "39.20"],
"en-ru": ["[36.4](http://matrix.statmt.org/matrix/output/1914?run_id=6724)", "33.47"],
"en-de": ["[43.1](http://matrix.statmt.org/matrix/output/1909?run_id=6862)", "42.83"],
"de-en": ["[42.3](http://matrix.statmt.org/matrix/output/1902?run_id=6750)", "41.35"],
}
A_ = f'''{src_lang}-{tgt_lang}'''
A_ = f'''
---
language:
- {src_lang}
- {tgt_lang}
thumbnail:
tags:
- translation
- wmt19
- facebook
license: apache-2.0
datasets:
- wmt19
metrics:
- bleu
---
# FSMT
## Model description
This is a ported version of [fairseq wmt19 transformer](https://github.com/pytorch/fairseq/blob/master/examples/wmt19/README.md) for {src_lang}-{tgt_lang}.
For more details, please see, [Facebook FAIR\'s WMT19 News Translation Task Submission](https://arxiv.org/abs/1907.06616).
The abbreviation FSMT stands for FairSeqMachineTranslation
All four models are available:
* [wmt19-en-ru](https://huggingface.co/facebook/wmt19-en-ru)
* [wmt19-ru-en](https://huggingface.co/facebook/wmt19-ru-en)
* [wmt19-en-de](https://huggingface.co/facebook/wmt19-en-de)
* [wmt19-de-en](https://huggingface.co/facebook/wmt19-de-en)
## Intended uses & limitations
#### How to use
```python
from transformers import FSMTForConditionalGeneration, FSMTTokenizer
mname = "facebook/wmt19-{src_lang}-{tgt_lang}"
tokenizer = FSMTTokenizer.from_pretrained(mname)
model = FSMTForConditionalGeneration.from_pretrained(mname)
input = "{texts[src_lang]}"
input_ids = tokenizer.encode(input, return_tensors="pt")
outputs = model.generate(input_ids)
decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(decoded) # {texts[tgt_lang]}
```
#### Limitations and bias
- The original (and this ported model) doesn\'t seem to handle well inputs with repeated sub-phrases, [content gets truncated](https://discuss.huggingface.co/t/issues-with-translating-inputs-containing-repeated-phrases/981)
## Training data
Pretrained weights were left identical to the original model released by fairseq. For more details, please, see the [paper](https://arxiv.org/abs/1907.06616).
## Eval results
pair | fairseq | transformers
-------|---------|----------
{pair} | {scores[pair][0]} | {scores[pair][1]}
The score is slightly below the score reported by `fairseq`, since `transformers`` currently doesn\'t support:
- model ensemble, therefore the best performing checkpoint was ported (``model4.pt``).
- re-ranking
The score was calculated using this code:
```bash
git clone https://github.com/huggingface/transformers
cd transformers
export PAIR={pair}
export DATA_DIR=data/$PAIR
export SAVE_DIR=data/$PAIR
export BS=8
export NUM_BEAMS=15
mkdir -p $DATA_DIR
sacrebleu -t wmt19 -l $PAIR --echo src > $DATA_DIR/val.source
sacrebleu -t wmt19 -l $PAIR --echo ref > $DATA_DIR/val.target
echo $PAIR
PYTHONPATH="src:examples/seq2seq" python examples/seq2seq/run_eval.py facebook/wmt19-$PAIR $DATA_DIR/val.source $SAVE_DIR/test_translations.txt --reference_path $DATA_DIR/val.target --score_path $SAVE_DIR/test_bleu.json --bs $BS --task translation --num_beams $NUM_BEAMS
```
note: fairseq reports using a beam of 50, so you should get a slightly higher score if re-run with `--num_beams 50`.
## Data Sources
- [training, etc.](http://www.statmt.org/wmt19/)
- [test set](http://matrix.statmt.org/test_sets/newstest2019.tgz?1556572561)
### BibTeX entry and citation info
```bibtex
@inproceedings{{...,
year={{2020}},
title={{Facebook FAIR\'s WMT19 News Translation Task Submission}},
author={{Ng, Nathan and Yee, Kyra and Baevski, Alexei and Ott, Myle and Auli, Michael and Edunov, Sergey}},
booktitle={{Proc. of WMT}},
}}
```
## TODO
- port model ensemble (fairseq uses 4 model checkpoints)
'''
os.makedirs(__UpperCamelCase ,exist_ok=__UpperCamelCase )
A_ = os.path.join(__UpperCamelCase ,"README.md" )
print(f'''Generating {path}''' )
with open(__UpperCamelCase ,"w" ,encoding="utf-8" ) as f:
f.write(__UpperCamelCase )
# make sure we are under the root of the project
__a :Optional[Any] = Path(__file__).resolve().parent.parent.parent
__a :Optional[Any] = repo_dir / 'model_cards'
for model_name in ["wmt19-ru-en", "wmt19-en-ru", "wmt19-en-de", "wmt19-de-en"]:
__a , __a , __a :int = model_name.split('-')
__a :str = model_cards_dir / 'facebook' / model_name
write_model_card(model_card_dir, src_lang=src_lang, tgt_lang=tgt_lang) | 86 | 1 |
from __future__ import annotations
def __snake_case ( __UpperCamelCase : list[int] ): # This function is recursive
"""simple docstring"""
A_ = len(__UpperCamelCase )
# If the array contains only one element, we return it (it's the stop condition of
# recursion)
if array_length <= 1:
return array
# Else
A_ = array[0]
A_ = False
A_ = 1
A_ = []
while not is_found and i < array_length:
if array[i] < pivot:
A_ = True
A_ = [element for element in array[i:] if element >= array[i]]
A_ = longest_subsequence(__UpperCamelCase )
if len(__UpperCamelCase ) > len(__UpperCamelCase ):
A_ = temp_array
else:
i += 1
A_ = [element for element in array[1:] if element >= pivot]
A_ = [pivot, *longest_subsequence(__UpperCamelCase )]
if len(__UpperCamelCase ) > len(__UpperCamelCase ):
return temp_array
else:
return longest_subseq
if __name__ == "__main__":
import doctest
doctest.testmod() | 86 |
from ..utils import DummyObject, requires_backends
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Optional[Any] = ['torch', 'transformers', 'onnx']
def __init__( self : str , *UpperCAmelCase : int , **UpperCAmelCase : List[Any] ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Dict , *UpperCAmelCase : Dict , **UpperCAmelCase : Union[str, Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[Any] , *UpperCAmelCase : List[Any] , **UpperCAmelCase : Union[str, Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : str = ['torch', 'transformers', 'onnx']
def __init__( self : Optional[int] , *UpperCAmelCase : List[str] , **UpperCAmelCase : int ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : List[str] , *UpperCAmelCase : List[str] , **UpperCAmelCase : List[Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[Any] , *UpperCAmelCase : Dict , **UpperCAmelCase : Optional[Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Optional[Any] = ['torch', 'transformers', 'onnx']
def __init__( self : Union[str, Any] , *UpperCAmelCase : Any , **UpperCAmelCase : Dict ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Tuple , *UpperCAmelCase : Dict , **UpperCAmelCase : str ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Dict , *UpperCAmelCase : List[str] , **UpperCAmelCase : List[str] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : int = ['torch', 'transformers', 'onnx']
def __init__( self : List[str] , *UpperCAmelCase : Dict , **UpperCAmelCase : Optional[Any] ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Any , *UpperCAmelCase : Union[str, Any] , **UpperCAmelCase : Dict ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[Any] , *UpperCAmelCase : int , **UpperCAmelCase : Optional[int] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Dict = ['torch', 'transformers', 'onnx']
def __init__( self : List[str] , *UpperCAmelCase : str , **UpperCAmelCase : int ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : List[str] , *UpperCAmelCase : Optional[int] , **UpperCAmelCase : Optional[int] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[int] , *UpperCAmelCase : int , **UpperCAmelCase : List[Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : int = ['torch', 'transformers', 'onnx']
def __init__( self : Tuple , *UpperCAmelCase : List[Any] , **UpperCAmelCase : Optional[Any] ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[int] , *UpperCAmelCase : Dict , **UpperCAmelCase : str ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[int] , *UpperCAmelCase : Optional[Any] , **UpperCAmelCase : int ):
requires_backends(cls , ["torch", "transformers", "onnx"] ) | 86 | 1 |
import unittest
from transformers import DebertaVaConfig, is_torch_available
from transformers.testing_utils import require_sentencepiece, require_tokenizers, require_torch, slow, torch_device
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from transformers import (
DebertaVaForMaskedLM,
DebertaVaForMultipleChoice,
DebertaVaForQuestionAnswering,
DebertaVaForSequenceClassification,
DebertaVaForTokenClassification,
DebertaVaModel,
)
from transformers.models.deberta_va.modeling_deberta_va import DEBERTA_V2_PRETRAINED_MODEL_ARCHIVE_LIST
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : List[Any] , UpperCAmelCase : Union[str, Any] , UpperCAmelCase : Any=13 , UpperCAmelCase : List[str]=7 , UpperCAmelCase : Dict=True , UpperCAmelCase : List[Any]=True , UpperCAmelCase : int=True , UpperCAmelCase : Union[str, Any]=True , UpperCAmelCase : List[Any]=99 , UpperCAmelCase : List[str]=32 , UpperCAmelCase : Tuple=5 , UpperCAmelCase : Union[str, Any]=4 , UpperCAmelCase : Dict=37 , UpperCAmelCase : str="gelu" , UpperCAmelCase : Tuple=0.1 , UpperCAmelCase : Any=0.1 , UpperCAmelCase : Tuple=512 , UpperCAmelCase : Dict=16 , UpperCAmelCase : int=2 , UpperCAmelCase : Any=0.02 , UpperCAmelCase : List[str]=False , UpperCAmelCase : List[str]=True , UpperCAmelCase : Tuple="None" , UpperCAmelCase : Optional[int]=3 , UpperCAmelCase : str=4 , UpperCAmelCase : List[str]=None , ):
A_ = parent
A_ = batch_size
A_ = seq_length
A_ = is_training
A_ = use_input_mask
A_ = use_token_type_ids
A_ = use_labels
A_ = vocab_size
A_ = hidden_size
A_ = num_hidden_layers
A_ = num_attention_heads
A_ = intermediate_size
A_ = hidden_act
A_ = hidden_dropout_prob
A_ = attention_probs_dropout_prob
A_ = max_position_embeddings
A_ = type_vocab_size
A_ = type_sequence_label_size
A_ = initializer_range
A_ = num_labels
A_ = num_choices
A_ = relative_attention
A_ = position_biased_input
A_ = pos_att_type
A_ = scope
def __A ( self : Tuple ):
A_ = ids_tensor([self.batch_size, self.seq_length] , self.vocab_size )
A_ = None
if self.use_input_mask:
A_ = ids_tensor([self.batch_size, self.seq_length] , vocab_size=2 )
A_ = None
if self.use_token_type_ids:
A_ = ids_tensor([self.batch_size, self.seq_length] , self.type_vocab_size )
A_ = None
A_ = None
A_ = None
if self.use_labels:
A_ = ids_tensor([self.batch_size] , self.type_sequence_label_size )
A_ = ids_tensor([self.batch_size, self.seq_length] , self.num_labels )
A_ = ids_tensor([self.batch_size] , self.num_choices )
A_ = self.get_config()
return config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
def __A ( self : Optional[int] ):
return DebertaVaConfig(
vocab_size=self.vocab_size , hidden_size=self.hidden_size , num_hidden_layers=self.num_hidden_layers , num_attention_heads=self.num_attention_heads , intermediate_size=self.intermediate_size , hidden_act=self.hidden_act , hidden_dropout_prob=self.hidden_dropout_prob , attention_probs_dropout_prob=self.attention_probs_dropout_prob , max_position_embeddings=self.max_position_embeddings , type_vocab_size=self.type_vocab_size , initializer_range=self.initializer_range , relative_attention=self.relative_attention , position_biased_input=self.position_biased_input , pos_att_type=self.pos_att_type , )
def __A ( self : Optional[int] , UpperCAmelCase : Tuple ):
self.parent.assertListEqual(list(result.loss.size() ) , [] )
def __A ( self : Tuple , UpperCAmelCase : List[Any] , UpperCAmelCase : Optional[int] , UpperCAmelCase : Union[str, Any] , UpperCAmelCase : List[str] , UpperCAmelCase : Optional[Any] , UpperCAmelCase : Union[str, Any] , UpperCAmelCase : List[str] ):
A_ = DebertaVaModel(config=UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
A_ = model(UpperCAmelCase , attention_mask=UpperCAmelCase , token_type_ids=UpperCAmelCase )[0]
A_ = model(UpperCAmelCase , token_type_ids=UpperCAmelCase )[0]
A_ = model(UpperCAmelCase )[0]
self.parent.assertListEqual(list(sequence_output.size() ) , [self.batch_size, self.seq_length, self.hidden_size] )
def __A ( self : str , UpperCAmelCase : List[Any] , UpperCAmelCase : int , UpperCAmelCase : Tuple , UpperCAmelCase : List[Any] , UpperCAmelCase : Union[str, Any] , UpperCAmelCase : Dict , UpperCAmelCase : Dict ):
A_ = DebertaVaForMaskedLM(config=UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
A_ = model(UpperCAmelCase , attention_mask=UpperCAmelCase , token_type_ids=UpperCAmelCase , labels=UpperCAmelCase )
self.parent.assertEqual(result.logits.shape , (self.batch_size, self.seq_length, self.vocab_size) )
def __A ( self : int , UpperCAmelCase : Dict , UpperCAmelCase : List[Any] , UpperCAmelCase : str , UpperCAmelCase : List[str] , UpperCAmelCase : int , UpperCAmelCase : Optional[Any] , UpperCAmelCase : Dict ):
A_ = self.num_labels
A_ = DebertaVaForSequenceClassification(UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
A_ = model(UpperCAmelCase , attention_mask=UpperCAmelCase , token_type_ids=UpperCAmelCase , labels=UpperCAmelCase )
self.parent.assertListEqual(list(result.logits.size() ) , [self.batch_size, self.num_labels] )
self.check_loss_output(UpperCAmelCase )
def __A ( self : Tuple , UpperCAmelCase : str , UpperCAmelCase : List[str] , UpperCAmelCase : Optional[Any] , UpperCAmelCase : str , UpperCAmelCase : Any , UpperCAmelCase : Optional[Any] , UpperCAmelCase : Dict ):
A_ = self.num_labels
A_ = DebertaVaForTokenClassification(config=UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
A_ = model(UpperCAmelCase , attention_mask=UpperCAmelCase , token_type_ids=UpperCAmelCase , labels=UpperCAmelCase )
self.parent.assertEqual(result.logits.shape , (self.batch_size, self.seq_length, self.num_labels) )
def __A ( self : Dict , UpperCAmelCase : Tuple , UpperCAmelCase : Optional[Any] , UpperCAmelCase : Optional[int] , UpperCAmelCase : Optional[Any] , UpperCAmelCase : List[str] , UpperCAmelCase : Optional[int] , UpperCAmelCase : Any ):
A_ = DebertaVaForQuestionAnswering(config=UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
A_ = model(
UpperCAmelCase , attention_mask=UpperCAmelCase , token_type_ids=UpperCAmelCase , start_positions=UpperCAmelCase , end_positions=UpperCAmelCase , )
self.parent.assertEqual(result.start_logits.shape , (self.batch_size, self.seq_length) )
self.parent.assertEqual(result.end_logits.shape , (self.batch_size, self.seq_length) )
def __A ( self : Dict , UpperCAmelCase : Dict , UpperCAmelCase : Optional[Any] , UpperCAmelCase : List[str] , UpperCAmelCase : Optional[Any] , UpperCAmelCase : List[str] , UpperCAmelCase : List[Any] , UpperCAmelCase : int ):
A_ = DebertaVaForMultipleChoice(config=UpperCAmelCase )
model.to(UpperCAmelCase )
model.eval()
A_ = input_ids.unsqueeze(1 ).expand(-1 , self.num_choices , -1 ).contiguous()
A_ = token_type_ids.unsqueeze(1 ).expand(-1 , self.num_choices , -1 ).contiguous()
A_ = input_mask.unsqueeze(1 ).expand(-1 , self.num_choices , -1 ).contiguous()
A_ = model(
UpperCAmelCase , attention_mask=UpperCAmelCase , token_type_ids=UpperCAmelCase , labels=UpperCAmelCase , )
self.parent.assertEqual(result.logits.shape , (self.batch_size, self.num_choices) )
def __A ( self : Optional[Any] ):
A_ = self.prepare_config_and_inputs()
(
(
A_
) , (
A_
) , (
A_
) , (
A_
) , (
A_
) , (
A_
) , (
A_
) ,
) = config_and_inputs
A_ = {"input_ids": input_ids, "token_type_ids": token_type_ids, "attention_mask": input_mask}
return config, inputs_dict
@require_torch
class _a ( snake_case_ , snake_case_ , unittest.TestCase ):
"""simple docstring"""
_lowerCamelCase : Dict = (
(
DebertaVaModel,
DebertaVaForMaskedLM,
DebertaVaForSequenceClassification,
DebertaVaForTokenClassification,
DebertaVaForQuestionAnswering,
DebertaVaForMultipleChoice,
)
if is_torch_available()
else ()
)
_lowerCamelCase : List[str] = (
{
'feature-extraction': DebertaVaModel,
'fill-mask': DebertaVaForMaskedLM,
'question-answering': DebertaVaForQuestionAnswering,
'text-classification': DebertaVaForSequenceClassification,
'token-classification': DebertaVaForTokenClassification,
'zero-shot': DebertaVaForSequenceClassification,
}
if is_torch_available()
else {}
)
_lowerCamelCase : Dict = True
_lowerCamelCase : Any = False
_lowerCamelCase : str = False
_lowerCamelCase : Optional[int] = False
_lowerCamelCase : Optional[Any] = False
def __A ( self : Dict ):
A_ = DebertaVaModelTester(self )
A_ = ConfigTester(self , config_class=UpperCAmelCase , hidden_size=37 )
def __A ( self : Union[str, Any] ):
self.config_tester.run_common_tests()
def __A ( self : List[str] ):
A_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_deberta_model(*UpperCAmelCase )
def __A ( self : Union[str, Any] ):
A_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_deberta_for_sequence_classification(*UpperCAmelCase )
def __A ( self : Any ):
A_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_deberta_for_masked_lm(*UpperCAmelCase )
def __A ( self : int ):
A_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_deberta_for_question_answering(*UpperCAmelCase )
def __A ( self : Any ):
A_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_deberta_for_token_classification(*UpperCAmelCase )
def __A ( self : str ):
A_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_deberta_for_multiple_choice(*UpperCAmelCase )
@slow
def __A ( self : Optional[int] ):
for model_name in DEBERTA_V2_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
A_ = DebertaVaModel.from_pretrained(UpperCAmelCase )
self.assertIsNotNone(UpperCAmelCase )
@require_torch
@require_sentencepiece
@require_tokenizers
class _a ( unittest.TestCase ):
"""simple docstring"""
@unittest.skip(reason="Model not available yet" )
def __A ( self : Dict ):
pass
@slow
def __A ( self : Tuple ):
A_ = DebertaVaModel.from_pretrained("microsoft/deberta-v2-xlarge" )
A_ = torch.tensor([[0, 31414, 232, 328, 740, 1140, 12695, 69, 46078, 1588, 2]] )
A_ = torch.tensor([[0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]] )
with torch.no_grad():
A_ = model(UpperCAmelCase , attention_mask=UpperCAmelCase )[0]
# compare the actual values for a slice.
A_ = torch.tensor(
[[[0.2_356, 0.1_948, 0.0_369], [-0.1_063, 0.3_586, -0.5_152], [-0.6_399, -0.0_259, -0.2_525]]] )
self.assertTrue(torch.allclose(output[:, 1:4, 1:4] , UpperCAmelCase , atol=1E-4 ) , f'''{output[:, 1:4, 1:4]}''' ) | 86 |
import torch
from diffusers import DDPMParallelScheduler
from .test_schedulers import SchedulerCommonTest
class _a ( snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Optional[Any] = (DDPMParallelScheduler,)
def __A ( self : List[Any] , **UpperCAmelCase : Optional[int] ):
A_ = {
"num_train_timesteps": 1000,
"beta_start": 0.0_001,
"beta_end": 0.02,
"beta_schedule": "linear",
"variance_type": "fixed_small",
"clip_sample": True,
}
config.update(**UpperCAmelCase )
return config
def __A ( self : Optional[Any] ):
for timesteps in [1, 5, 100, 1000]:
self.check_over_configs(num_train_timesteps=UpperCAmelCase )
def __A ( self : Dict ):
for beta_start, beta_end in zip([0.0_001, 0.001, 0.01, 0.1] , [0.002, 0.02, 0.2, 2] ):
self.check_over_configs(beta_start=UpperCAmelCase , beta_end=UpperCAmelCase )
def __A ( self : int ):
for schedule in ["linear", "squaredcos_cap_v2"]:
self.check_over_configs(beta_schedule=UpperCAmelCase )
def __A ( self : Tuple ):
for variance in ["fixed_small", "fixed_large", "other"]:
self.check_over_configs(variance_type=UpperCAmelCase )
def __A ( self : int ):
for clip_sample in [True, False]:
self.check_over_configs(clip_sample=UpperCAmelCase )
def __A ( self : Union[str, Any] ):
self.check_over_configs(thresholding=UpperCAmelCase )
for threshold in [0.5, 1.0, 2.0]:
for prediction_type in ["epsilon", "sample", "v_prediction"]:
self.check_over_configs(
thresholding=UpperCAmelCase , prediction_type=UpperCAmelCase , sample_max_value=UpperCAmelCase , )
def __A ( self : Optional[int] ):
for prediction_type in ["epsilon", "sample", "v_prediction"]:
self.check_over_configs(prediction_type=UpperCAmelCase )
def __A ( self : Tuple ):
for t in [0, 500, 999]:
self.check_over_forward(time_step=UpperCAmelCase )
def __A ( self : Tuple ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
assert torch.sum(torch.abs(scheduler._get_variance(0 ) - 0.0 ) ) < 1E-5
assert torch.sum(torch.abs(scheduler._get_variance(487 ) - 0.00_979 ) ) < 1E-5
assert torch.sum(torch.abs(scheduler._get_variance(999 ) - 0.02 ) ) < 1E-5
def __A ( self : List[Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = len(UpperCAmelCase )
A_ = self.dummy_model()
A_ = self.dummy_sample_deter
A_ = self.dummy_sample_deter + 0.1
A_ = self.dummy_sample_deter - 0.1
A_ = samplea.shape[0]
A_ = torch.stack([samplea, samplea, samplea] , dim=0 )
A_ = torch.arange(UpperCAmelCase )[0:3, None].repeat(1 , UpperCAmelCase )
A_ = model(samples.flatten(0 , 1 ) , timesteps.flatten(0 , 1 ) )
A_ = scheduler.batch_step_no_noise(UpperCAmelCase , timesteps.flatten(0 , 1 ) , samples.flatten(0 , 1 ) )
A_ = torch.sum(torch.abs(UpperCAmelCase ) )
A_ = torch.mean(torch.abs(UpperCAmelCase ) )
assert abs(result_sum.item() - 1_153.1_833 ) < 1E-2
assert abs(result_mean.item() - 0.5_005 ) < 1E-3
def __A ( self : Tuple ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = len(UpperCAmelCase )
A_ = self.dummy_model()
A_ = self.dummy_sample_deter
A_ = torch.manual_seed(0 )
for t in reversed(range(UpperCAmelCase ) ):
# 1. predict noise residual
A_ = model(UpperCAmelCase , UpperCAmelCase )
# 2. predict previous mean of sample x_t-1
A_ = scheduler.step(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase , generator=UpperCAmelCase ).prev_sample
A_ = pred_prev_sample
A_ = torch.sum(torch.abs(UpperCAmelCase ) )
A_ = torch.mean(torch.abs(UpperCAmelCase ) )
assert abs(result_sum.item() - 258.9_606 ) < 1E-2
assert abs(result_mean.item() - 0.3_372 ) < 1E-3
def __A ( self : Tuple ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config(prediction_type="v_prediction" )
A_ = scheduler_class(**UpperCAmelCase )
A_ = len(UpperCAmelCase )
A_ = self.dummy_model()
A_ = self.dummy_sample_deter
A_ = torch.manual_seed(0 )
for t in reversed(range(UpperCAmelCase ) ):
# 1. predict noise residual
A_ = model(UpperCAmelCase , UpperCAmelCase )
# 2. predict previous mean of sample x_t-1
A_ = scheduler.step(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase , generator=UpperCAmelCase ).prev_sample
A_ = pred_prev_sample
A_ = torch.sum(torch.abs(UpperCAmelCase ) )
A_ = torch.mean(torch.abs(UpperCAmelCase ) )
assert abs(result_sum.item() - 202.0_296 ) < 1E-2
assert abs(result_mean.item() - 0.2_631 ) < 1E-3
def __A ( self : Union[str, Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = [100, 87, 50, 1, 0]
scheduler.set_timesteps(timesteps=UpperCAmelCase )
A_ = scheduler.timesteps
for i, timestep in enumerate(UpperCAmelCase ):
if i == len(UpperCAmelCase ) - 1:
A_ = -1
else:
A_ = timesteps[i + 1]
A_ = scheduler.previous_timestep(UpperCAmelCase )
A_ = prev_t.item()
self.assertEqual(UpperCAmelCase , UpperCAmelCase )
def __A ( self : List[Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = [100, 87, 50, 51, 0]
with self.assertRaises(UpperCAmelCase , msg="`custom_timesteps` must be in descending order." ):
scheduler.set_timesteps(timesteps=UpperCAmelCase )
def __A ( self : List[Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = [100, 87, 50, 1, 0]
A_ = len(UpperCAmelCase )
with self.assertRaises(UpperCAmelCase , msg="Can only pass one of `num_inference_steps` or `custom_timesteps`." ):
scheduler.set_timesteps(num_inference_steps=UpperCAmelCase , timesteps=UpperCAmelCase )
def __A ( self : Optional[Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = [scheduler.config.num_train_timesteps]
with self.assertRaises(
UpperCAmelCase , msg="`timesteps` must start before `self.config.train_timesteps`: {scheduler.config.num_train_timesteps}}" , ):
scheduler.set_timesteps(timesteps=UpperCAmelCase ) | 86 | 1 |
# Copyright (c) 2021-, NVIDIA CORPORATION. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
####################################################################################################
#
# Note: If when running this conversion script you're getting an exception:
# ModuleNotFoundError: No module named 'megatron.model.enums'
# you need to tell python where to find the clone of Megatron-LM, e.g.:
#
# cd /tmp
# git clone https://github.com/NVIDIA/Megatron-LM
# PYTHONPATH=/tmp/Megatron-LM python src/transformers/models/megatron_gpt2/convert_megatron_gpt2_checkpoint.py ...
#
# if you already have it cloned elsewhere, simply adjust the path to the existing path
#
# If the training was done using a Megatron-LM fork, e.g.,
# https://github.com/microsoft/Megatron-DeepSpeed/ then chances are that you need to have that one
# in your path, i.e., /path/to/Megatron-DeepSpeed/
#
import argparse
import os
import re
import zipfile
import torch
from transformers import AutoTokenizer, GPTaConfig
def __snake_case ( __UpperCamelCase : Tuple ,__UpperCamelCase : Dict ,__UpperCamelCase : Tuple=0 ):
"""simple docstring"""
if name is None:
A_ = None
else:
A_ = "." * max(0 ,spaces - 2 ) + "# {:" + str(50 - spaces ) + "s}"
A_ = fmt.format(__UpperCamelCase )
# Print and recurse (if needed).
if isinstance(__UpperCamelCase ,__UpperCamelCase ):
if msg is not None:
print(__UpperCamelCase )
for k in val.keys():
recursive_print(__UpperCamelCase ,val[k] ,spaces + 2 )
elif isinstance(__UpperCamelCase ,torch.Tensor ):
print(__UpperCamelCase ,":" ,val.size() )
else:
print(__UpperCamelCase ,":" ,__UpperCamelCase )
def __snake_case ( __UpperCamelCase : List[str] ,__UpperCamelCase : Tuple ,__UpperCamelCase : Union[str, Any] ,__UpperCamelCase : Any ,__UpperCamelCase : Any ):
"""simple docstring"""
A_ = param.size()
if checkpoint_version == 1.0:
# version 1.0 stores [num_heads * hidden_size * num_splits, :]
A_ = (num_heads, hidden_size, num_splits) + input_shape[1:]
A_ = param.view(*__UpperCamelCase )
A_ = param.transpose(0 ,2 )
A_ = param.transpose(1 ,2 ).contiguous()
elif checkpoint_version >= 2.0:
# other versions store [num_heads * num_splits * hidden_size, :]
A_ = (num_heads, num_splits, hidden_size) + input_shape[1:]
A_ = param.view(*__UpperCamelCase )
A_ = param.transpose(0 ,1 ).contiguous()
A_ = param.view(*__UpperCamelCase )
return param
def __snake_case ( __UpperCamelCase : str ,__UpperCamelCase : Any ,__UpperCamelCase : Any ):
"""simple docstring"""
A_ = {}
# old versions did not store training args
A_ = input_state_dict.get("args" ,__UpperCamelCase )
if ds_args is not None:
# do not make the user write a config file when the exact dimensions/sizes are already in the checkpoint
# from pprint import pprint
# pprint(vars(ds_args))
A_ = ds_args.padded_vocab_size
A_ = ds_args.max_position_embeddings
A_ = ds_args.hidden_size
A_ = ds_args.num_layers
A_ = ds_args.num_attention_heads
A_ = ds_args.ffn_hidden_size
# pprint(config)
# The number of heads.
A_ = config.n_head
# The hidden_size per head.
A_ = config.n_embd // config.n_head
# Megatron-LM checkpoint version
if "checkpoint_version" in input_state_dict.keys():
A_ = input_state_dict["checkpoint_version"]
else:
A_ = 0.0
# The model.
A_ = input_state_dict["model"]
# The language model.
A_ = model["language_model"]
# The embeddings.
A_ = lm["embedding"]
# The word embeddings.
A_ = embeddings["word_embeddings"]["weight"]
# Truncate the embedding table to vocab_size rows.
A_ = word_embeddings[: config.vocab_size, :]
A_ = word_embeddings
# The position embeddings.
A_ = embeddings["position_embeddings"]["weight"]
# Read the causal mask dimension (seqlen). [max_sequence_length, hidden_size]
A_ = pos_embeddings.size(0 )
if n_positions != config.n_positions:
raise ValueError(
f'''pos_embeddings.max_sequence_length={n_positions} and config.n_positions={config.n_positions} don\'t match''' )
# Store the position embeddings.
A_ = pos_embeddings
# The transformer.
A_ = lm["transformer"] if "transformer" in lm.keys() else lm["encoder"]
# The regex to extract layer names.
A_ = re.compile(R"layers\.(\d+)\.([a-z0-9_.]+)\.([a-z]+)" )
# The simple map of names for "automated" rules.
A_ = {
"attention.dense": ".attn.c_proj.",
"self_attention.dense": ".attn.c_proj.",
"mlp.dense_h_to_4h": ".mlp.c_fc.",
"mlp.dense_4h_to_h": ".mlp.c_proj.",
}
# Extract the layers.
for key, val in transformer.items():
# Match the name.
A_ = layer_re.match(__UpperCamelCase )
# Stop if that's not a layer
if m is None:
break
# The index of the layer.
A_ = int(m.group(1 ) )
# The name of the operation.
A_ = m.group(2 )
# Is it a weight or a bias?
A_ = m.group(3 )
# The name of the layer.
A_ = f'''transformer.h.{layer_idx}'''
# For layernorm(s), simply store the layer norm.
if op_name.endswith("layernorm" ):
A_ = "ln_1" if op_name.startswith("input" ) else "ln_2"
A_ = val
# Transpose the QKV matrix.
elif (
op_name == "attention.query_key_value" or op_name == "self_attention.query_key_value"
) and weight_or_bias == "weight":
# Insert a tensor of 1x1xDxD bias.
A_ = torch.tril(torch.ones((n_positions, n_positions) ,dtype=torch.floataa ) ).view(
1 ,1 ,__UpperCamelCase ,__UpperCamelCase )
A_ = causal_mask
# Insert a "dummy" tensor for masked_bias.
A_ = torch.tensor(-1E4 ,dtype=torch.floataa )
A_ = masked_bias
A_ = fix_query_key_value_ordering(__UpperCamelCase ,__UpperCamelCase ,3 ,__UpperCamelCase ,__UpperCamelCase )
# Megatron stores (3*D) x D but transformers-GPT2 expects D x 3*D.
A_ = out_val.transpose(0 ,1 ).contiguous()
# Store.
A_ = out_val
# Transpose the bias.
elif (
op_name == "attention.query_key_value" or op_name == "self_attention.query_key_value"
) and weight_or_bias == "bias":
A_ = fix_query_key_value_ordering(__UpperCamelCase ,__UpperCamelCase ,3 ,__UpperCamelCase ,__UpperCamelCase )
# Store. No change of shape.
A_ = out_val
# Transpose the weights.
elif weight_or_bias == "weight":
A_ = megatron_to_transformers[op_name]
A_ = val.transpose(0 ,1 )
# Copy the bias.
elif weight_or_bias == "bias":
A_ = megatron_to_transformers[op_name]
A_ = val
# DEBUG.
assert config.n_layer == layer_idx + 1
# The final layernorm.
A_ = transformer["final_layernorm.weight"]
A_ = transformer["final_layernorm.bias"]
# For LM head, transformers' wants the matrix to weight embeddings.
A_ = word_embeddings
# It should be done!
return output_state_dict
def __snake_case ( ):
"""simple docstring"""
A_ = argparse.ArgumentParser()
parser.add_argument("--print-checkpoint-structure" ,action="store_true" )
parser.add_argument(
"path_to_checkpoint" ,type=__UpperCamelCase ,help="Path to the checkpoint file (.zip archive or direct .pt file)" ,)
parser.add_argument(
"--config_file" ,default="" ,type=__UpperCamelCase ,help="An optional config json file describing the pre-trained model." ,)
A_ = parser.parse_args()
# Extract the basename.
A_ = os.path.dirname(args.path_to_checkpoint )
# Load the model.
# the .zip is very optional, let's keep it for backward compatibility
print(f'''Extracting PyTorch state dictionary from {args.path_to_checkpoint}''' )
if args.path_to_checkpoint.endswith(".zip" ):
with zipfile.ZipFile(args.path_to_checkpoint ,"r" ) as checkpoint:
with checkpoint.open("release/mp_rank_00/model_optim_rng.pt" ) as pytorch_dict:
A_ = torch.load(__UpperCamelCase ,map_location="cpu" )
else:
A_ = torch.load(args.path_to_checkpoint ,map_location="cpu" )
A_ = input_state_dict.get("args" ,__UpperCamelCase )
# Read the config, or default to the model released by NVIDIA.
if args.config_file == "":
if ds_args is not None:
if ds_args.bias_gelu_fusion:
A_ = "gelu_fast"
elif ds_args.openai_gelu:
A_ = "gelu_new"
else:
A_ = "gelu"
else:
# in the very early days this used to be "gelu_new"
A_ = "gelu_new"
# Spell out all parameters in case the defaults change.
A_ = GPTaConfig(
vocab_size=5_0257 ,n_positions=1024 ,n_embd=1024 ,n_layer=24 ,n_head=16 ,n_inner=4096 ,activation_function=__UpperCamelCase ,resid_pdrop=0.1 ,embd_pdrop=0.1 ,attn_pdrop=0.1 ,layer_norm_epsilon=1E-5 ,initializer_range=0.02 ,summary_type="cls_index" ,summary_use_proj=__UpperCamelCase ,summary_activation=__UpperCamelCase ,summary_proj_to_labels=__UpperCamelCase ,summary_first_dropout=0.1 ,scale_attn_weights=__UpperCamelCase ,use_cache=__UpperCamelCase ,bos_token_id=5_0256 ,eos_token_id=5_0256 ,)
else:
A_ = GPTaConfig.from_json_file(args.config_file )
A_ = ["GPT2LMHeadModel"]
# Convert.
print("Converting" )
A_ = convert_megatron_checkpoint(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
# Print the structure of converted state dict.
if args.print_checkpoint_structure:
recursive_print(__UpperCamelCase ,__UpperCamelCase )
# Add tokenizer class info to config
# see https://github.com/huggingface/transformers/issues/13906)
if ds_args is not None:
A_ = ds_args.tokenizer_type
if tokenizer_type == "GPT2BPETokenizer":
A_ = "gpt2"
elif tokenizer_type == "PretrainedFromHF":
A_ = ds_args.tokenizer_name_or_path
else:
raise ValueError(f'''Unrecognized tokenizer_type {tokenizer_type}''' )
else:
A_ = "gpt2"
A_ = AutoTokenizer.from_pretrained(__UpperCamelCase )
A_ = type(__UpperCamelCase ).__name__
A_ = tokenizer_class
# Store the config to file.
print("Saving config" )
config.save_pretrained(__UpperCamelCase )
# Save tokenizer based on args
print(f'''Adding {tokenizer_class} tokenizer files''' )
tokenizer.save_pretrained(__UpperCamelCase )
# Store the state_dict to file.
A_ = os.path.join(__UpperCamelCase ,"pytorch_model.bin" )
print(f'''Saving checkpoint to "{output_checkpoint_file}"''' )
torch.save(__UpperCamelCase ,__UpperCamelCase )
####################################################################################################
if __name__ == "__main__":
main()
#################################################################################################### | 86 |
import argparse
import json
import os
import torch
from transformers import LukeConfig, LukeModel, LukeTokenizer, RobertaTokenizer
from transformers.tokenization_utils_base import AddedToken
@torch.no_grad()
def __snake_case ( __UpperCamelCase : Tuple ,__UpperCamelCase : Dict ,__UpperCamelCase : Union[str, Any] ,__UpperCamelCase : List[Any] ,__UpperCamelCase : List[Any] ):
"""simple docstring"""
with open(__UpperCamelCase ) as metadata_file:
A_ = json.load(__UpperCamelCase )
A_ = LukeConfig(use_entity_aware_attention=__UpperCamelCase ,**metadata["model_config"] )
# Load in the weights from the checkpoint_path
A_ = torch.load(__UpperCamelCase ,map_location="cpu" )
# Load the entity vocab file
A_ = load_entity_vocab(__UpperCamelCase )
A_ = RobertaTokenizer.from_pretrained(metadata["model_config"]["bert_model_name"] )
# Add special tokens to the token vocabulary for downstream tasks
A_ = AddedToken("<ent>" ,lstrip=__UpperCamelCase ,rstrip=__UpperCamelCase )
A_ = AddedToken("<ent2>" ,lstrip=__UpperCamelCase ,rstrip=__UpperCamelCase )
tokenizer.add_special_tokens({"additional_special_tokens": [entity_token_a, entity_token_a]} )
config.vocab_size += 2
print(f'''Saving tokenizer to {pytorch_dump_folder_path}''' )
tokenizer.save_pretrained(__UpperCamelCase )
with open(os.path.join(__UpperCamelCase ,LukeTokenizer.vocab_files_names["entity_vocab_file"] ) ,"w" ) as f:
json.dump(__UpperCamelCase ,__UpperCamelCase )
A_ = LukeTokenizer.from_pretrained(__UpperCamelCase )
# Initialize the embeddings of the special tokens
A_ = state_dict["embeddings.word_embeddings.weight"]
A_ = word_emb[tokenizer.convert_tokens_to_ids(["@"] )[0]].unsqueeze(0 )
A_ = word_emb[tokenizer.convert_tokens_to_ids(["#"] )[0]].unsqueeze(0 )
A_ = torch.cat([word_emb, ent_emb, enta_emb] )
# Initialize the query layers of the entity-aware self-attention mechanism
for layer_index in range(config.num_hidden_layers ):
for matrix_name in ["query.weight", "query.bias"]:
A_ = f'''encoder.layer.{layer_index}.attention.self.'''
A_ = state_dict[prefix + matrix_name]
A_ = state_dict[prefix + matrix_name]
A_ = state_dict[prefix + matrix_name]
# Initialize the embedding of the [MASK2] entity using that of the [MASK] entity for downstream tasks
A_ = state_dict["entity_embeddings.entity_embeddings.weight"]
A_ = entity_emb[entity_vocab["[MASK]"]]
A_ = LukeModel(config=__UpperCamelCase ).eval()
A_ , A_ = model.load_state_dict(__UpperCamelCase ,strict=__UpperCamelCase )
if not (len(__UpperCamelCase ) == 1 and missing_keys[0] == "embeddings.position_ids"):
raise ValueError(f'''Missing keys {", ".join(__UpperCamelCase )}. Expected only missing embeddings.position_ids''' )
if not (all(key.startswith("entity_predictions" ) or key.startswith("lm_head" ) for key in unexpected_keys )):
raise ValueError(
"Unexpected keys"
f''' {", ".join([key for key in unexpected_keys if not (key.startswith("entity_predictions" ) or key.startswith("lm_head" ))] )}''' )
# Check outputs
A_ = LukeTokenizer.from_pretrained(__UpperCamelCase ,task="entity_classification" )
A_ = (
"Top seed Ana Ivanovic said on Thursday she could hardly believe her luck as a fortuitous netcord helped the"
" new world number one avoid a humiliating second- round exit at Wimbledon ."
)
A_ = (39, 42)
A_ = tokenizer(__UpperCamelCase ,entity_spans=[span] ,add_prefix_space=__UpperCamelCase ,return_tensors="pt" )
A_ = model(**__UpperCamelCase )
# Verify word hidden states
if model_size == "large":
A_ = torch.Size((1, 42, 1024) )
A_ = torch.tensor(
[[0.0133, 0.0865, 0.0095], [0.3093, -0.2576, -0.7418], [-0.1720, -0.2117, -0.2869]] )
else: # base
A_ = torch.Size((1, 42, 768) )
A_ = torch.tensor([[0.0037, 0.1368, -0.0091], [0.1099, 0.3329, -0.1095], [0.0765, 0.5335, 0.1179]] )
if not (outputs.last_hidden_state.shape == expected_shape):
raise ValueError(
f'''Outputs.last_hidden_state.shape is {outputs.last_hidden_state.shape}, Expected shape is {expected_shape}''' )
if not torch.allclose(outputs.last_hidden_state[0, :3, :3] ,__UpperCamelCase ,atol=1E-4 ):
raise ValueError
# Verify entity hidden states
if model_size == "large":
A_ = torch.Size((1, 1, 1024) )
A_ = torch.tensor([[0.0466, -0.0106, -0.0179]] )
else: # base
A_ = torch.Size((1, 1, 768) )
A_ = torch.tensor([[0.1457, 0.1044, 0.0174]] )
if not (outputs.entity_last_hidden_state.shape != expected_shape):
raise ValueError(
f'''Outputs.entity_last_hidden_state.shape is {outputs.entity_last_hidden_state.shape}, Expected shape is'''
f''' {expected_shape}''' )
if not torch.allclose(outputs.entity_last_hidden_state[0, :3, :3] ,__UpperCamelCase ,atol=1E-4 ):
raise ValueError
# Finally, save our PyTorch model and tokenizer
print("Saving PyTorch model to {}".format(__UpperCamelCase ) )
model.save_pretrained(__UpperCamelCase )
def __snake_case ( __UpperCamelCase : str ):
"""simple docstring"""
A_ = {}
with open(__UpperCamelCase ,"r" ,encoding="utf-8" ) as f:
for index, line in enumerate(__UpperCamelCase ):
A_ , A_ = line.rstrip().split("\t" )
A_ = index
return entity_vocab
if __name__ == "__main__":
__a :Dict = argparse.ArgumentParser()
# Required parameters
parser.add_argument('--checkpoint_path', type=str, help='Path to a pytorch_model.bin file.')
parser.add_argument(
'--metadata_path', default=None, type=str, help='Path to a metadata.json file, defining the configuration.'
)
parser.add_argument(
'--entity_vocab_path',
default=None,
type=str,
help='Path to an entity_vocab.tsv file, containing the entity vocabulary.',
)
parser.add_argument(
'--pytorch_dump_folder_path', default=None, type=str, help='Path to where to dump the output PyTorch model.'
)
parser.add_argument(
'--model_size', default='base', type=str, choices=['base', 'large'], help='Size of the model to be converted.'
)
__a :Tuple = parser.parse_args()
convert_luke_checkpoint(
args.checkpoint_path,
args.metadata_path,
args.entity_vocab_path,
args.pytorch_dump_folder_path,
args.model_size,
) | 86 | 1 |
import pickle
import shutil
import tempfile
import unittest
from transformers import SPIECE_UNDERLINE, XGLMTokenizer, XGLMTokenizerFast
from transformers.testing_utils import get_tests_dir, require_sentencepiece, require_tokenizers, slow
from transformers.utils import cached_property
from ...test_tokenization_common import TokenizerTesterMixin
__a :Optional[Any] = get_tests_dir('fixtures/test_sentencepiece.model')
@require_sentencepiece
@require_tokenizers
class _a ( snake_case_ , unittest.TestCase ):
"""simple docstring"""
_lowerCamelCase : int = XGLMTokenizer
_lowerCamelCase : Union[str, Any] = XGLMTokenizerFast
_lowerCamelCase : Dict = True
_lowerCamelCase : Optional[Any] = True
def __A ( self : Tuple ):
super().setUp()
# We have a SentencePiece fixture for testing
A_ = XGLMTokenizer(UpperCAmelCase , keep_accents=UpperCAmelCase )
tokenizer.save_pretrained(self.tmpdirname )
def __A ( self : Dict ):
A_ = "<pad>"
A_ = 1
self.assertEqual(self.get_tokenizer()._convert_token_to_id(UpperCAmelCase ) , UpperCAmelCase )
self.assertEqual(self.get_tokenizer()._convert_id_to_token(UpperCAmelCase ) , UpperCAmelCase )
def __A ( self : Optional[int] ):
A_ = list(self.get_tokenizer().get_vocab().keys() )
self.assertEqual(vocab_keys[0] , "<s>" )
self.assertEqual(vocab_keys[1] , "<pad>" )
self.assertEqual(len(UpperCAmelCase ) , 1008 )
def __A ( self : int ):
self.assertEqual(self.get_tokenizer().vocab_size , 1008 )
def __A ( self : Dict ):
A_ = XGLMTokenizer(UpperCAmelCase , keep_accents=UpperCAmelCase )
A_ = tokenizer.tokenize("This is a test" )
self.assertListEqual(UpperCAmelCase , ["▁This", "▁is", "▁a", "▁t", "est"] )
self.assertListEqual(
tokenizer.convert_tokens_to_ids(UpperCAmelCase ) , [value + tokenizer.fairseq_offset for value in [285, 46, 10, 170, 382]] , )
A_ = tokenizer.tokenize("I was born in 92000, and this is falsé." )
self.assertListEqual(
UpperCAmelCase , [
SPIECE_UNDERLINE + "I",
SPIECE_UNDERLINE + "was",
SPIECE_UNDERLINE + "b",
"or",
"n",
SPIECE_UNDERLINE + "in",
SPIECE_UNDERLINE + "",
"9",
"2",
"0",
"0",
"0",
",",
SPIECE_UNDERLINE + "and",
SPIECE_UNDERLINE + "this",
SPIECE_UNDERLINE + "is",
SPIECE_UNDERLINE + "f",
"al",
"s",
"é",
".",
] , )
A_ = tokenizer.convert_tokens_to_ids(UpperCAmelCase )
self.assertListEqual(
UpperCAmelCase , [
value + tokenizer.fairseq_offset
for value in [8, 21, 84, 55, 24, 19, 7, 2, 602, 347, 347, 347, 3, 12, 66, 46, 72, 80, 6, 2, 4]
] , )
A_ = tokenizer.convert_ids_to_tokens(UpperCAmelCase )
self.assertListEqual(
UpperCAmelCase , [
SPIECE_UNDERLINE + "I",
SPIECE_UNDERLINE + "was",
SPIECE_UNDERLINE + "b",
"or",
"n",
SPIECE_UNDERLINE + "in",
SPIECE_UNDERLINE + "",
"<unk>",
"2",
"0",
"0",
"0",
",",
SPIECE_UNDERLINE + "and",
SPIECE_UNDERLINE + "this",
SPIECE_UNDERLINE + "is",
SPIECE_UNDERLINE + "f",
"al",
"s",
"<unk>",
".",
] , )
@cached_property
def __A ( self : List[Any] ):
return XGLMTokenizer.from_pretrained("facebook/xglm-564M" )
def __A ( self : Union[str, Any] ):
with tempfile.NamedTemporaryFile() as f:
shutil.copyfile(UpperCAmelCase , f.name )
A_ = XGLMTokenizer(f.name , keep_accents=UpperCAmelCase )
A_ = pickle.dumps(UpperCAmelCase )
pickle.loads(UpperCAmelCase )
def __A ( self : str ):
if not self.test_rust_tokenizer:
return
A_ = self.get_tokenizer()
A_ = self.get_rust_tokenizer()
A_ = "I was born in 92000, and this is falsé."
A_ = tokenizer.tokenize(UpperCAmelCase )
A_ = rust_tokenizer.tokenize(UpperCAmelCase )
self.assertListEqual(UpperCAmelCase , UpperCAmelCase )
A_ = tokenizer.encode(UpperCAmelCase , add_special_tokens=UpperCAmelCase )
A_ = rust_tokenizer.encode(UpperCAmelCase , add_special_tokens=UpperCAmelCase )
self.assertListEqual(UpperCAmelCase , UpperCAmelCase )
A_ = self.get_rust_tokenizer()
A_ = tokenizer.encode(UpperCAmelCase )
A_ = rust_tokenizer.encode(UpperCAmelCase )
self.assertListEqual(UpperCAmelCase , UpperCAmelCase )
@slow
def __A ( self : Optional[int] ):
A_ = "Hello World!"
A_ = [2, 31227, 4447, 35]
self.assertListEqual(UpperCAmelCase , self.big_tokenizer.encode(UpperCAmelCase ) )
@slow
def __A ( self : Any ):
A_ = (
"This is a very long text with a lot of weird characters, such as: . , ~ ? ( ) \" [ ] ! : - . Also we will"
" add words that should not exsist and be tokenized to unk, such as saoneuhaoesuth"
)
# fmt: off
A_ = [2, 1018, 67, 11, 1988, 2617, 5631, 278, 11, 3407, 48, 71630, 28085, 4, 3234, 157, 13, 6, 5, 6, 4, 3526, 768, 15, 659, 57, 298, 3983, 864, 129, 21, 6, 5, 13675, 377, 652, 7580, 10341, 155, 2817, 422, 1666, 7, 1674, 53, 113, 202277, 17892, 33, 60, 87, 4, 3234, 157, 61, 2667, 52376, 19, 88, 23, 735]
# fmt: on
self.assertListEqual(UpperCAmelCase , self.big_tokenizer.encode(UpperCAmelCase ) )
@slow
def __A ( self : Any ):
# fmt: off
A_ = {
"input_ids": [[2, 108825, 1163, 15, 88010, 473, 15898, 157, 13672, 1857, 312, 8, 238021, 1163, 53, 13672, 1857, 312, 8, 53283, 182396, 8, 18566, 16, 36733, 4101, 8, 230, 244017, 122553, 7, 15, 132597, 4, 293, 12511, 7610, 4, 3414, 132597, 9, 4, 32361, 362, 4, 734, 28512, 32569, 18, 4, 32361, 26096, 14982, 73, 18715, 21433, 235261, 15, 492, 12427, 16, 53, 18715, 21433, 65454, 15, 23659, 563, 16, 278, 597, 2843, 595, 7931, 182396, 64186, 22, 886, 595, 132981, 53, 25540, 3449, 43982, 39901, 5951, 878, 330, 4, 27694, 80269, 312, 53, 6517, 11780, 611, 20408, 5], [2, 6, 132597, 67, 42897, 33, 592, 8, 163729, 25540, 361, 136997, 109514, 173230, 7, 501, 60, 102913, 196, 5631, 235, 63243, 473, 6, 231757, 74, 5277, 7905, 53, 3095, 37317, 22, 454, 183874, 5], [2, 268, 31298, 46530, 6, 132935, 43831, 7, 597, 32, 24, 3688, 9865, 5]],
"attention_mask": [[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]]
} # noqa: E501
# fmt: on
self.tokenizer_integration_test_util(
expected_encoding=UpperCAmelCase , model_name="facebook/xglm-564M" , padding=UpperCAmelCase , ) | 86 |
import math
import os
from copy import deepcopy
import datasets
import evaluate
import torch
import transformers
from datasets import load_dataset
from torch.utils.data import DataLoader
from transformers import AutoModelForSequenceClassification, AutoTokenizer
from accelerate import Accelerator
from accelerate.test_utils import RegressionDataset, RegressionModel
from accelerate.utils import is_tpu_available, set_seed
__a :Optional[Any] = 'true'
def __snake_case ( __UpperCamelCase : Any ,__UpperCamelCase : List[Any]=82 ,__UpperCamelCase : Dict=16 ):
"""simple docstring"""
set_seed(42 )
A_ = RegressionModel()
A_ = deepcopy(__UpperCamelCase )
A_ = RegressionDataset(length=__UpperCamelCase )
A_ = DataLoader(__UpperCamelCase ,batch_size=__UpperCamelCase )
model.to(accelerator.device )
A_ , A_ = accelerator.prepare(__UpperCamelCase ,__UpperCamelCase )
return model, ddp_model, dataloader
def __snake_case ( __UpperCamelCase : Accelerator ,__UpperCamelCase : Dict=False ):
"""simple docstring"""
A_ = AutoTokenizer.from_pretrained("hf-internal-testing/mrpc-bert-base-cased" )
A_ = load_dataset("glue" ,"mrpc" ,split="validation" )
def tokenize_function(__UpperCamelCase : Optional[Any] ):
A_ = tokenizer(examples["sentence1"] ,examples["sentence2"] ,truncation=__UpperCamelCase ,max_length=__UpperCamelCase )
return outputs
with accelerator.main_process_first():
A_ = dataset.map(
__UpperCamelCase ,batched=__UpperCamelCase ,remove_columns=["idx", "sentence1", "sentence2"] ,)
A_ = tokenized_datasets.rename_column("label" ,"labels" )
def collate_fn(__UpperCamelCase : Union[str, Any] ):
if use_longest:
return tokenizer.pad(__UpperCamelCase ,padding="longest" ,return_tensors="pt" )
return tokenizer.pad(__UpperCamelCase ,padding="max_length" ,max_length=128 ,return_tensors="pt" )
return DataLoader(__UpperCamelCase ,shuffle=__UpperCamelCase ,collate_fn=__UpperCamelCase ,batch_size=16 )
def __snake_case ( __UpperCamelCase : Optional[Any] ,__UpperCamelCase : str ):
"""simple docstring"""
A_ = Accelerator(dispatch_batches=__UpperCamelCase ,split_batches=__UpperCamelCase )
A_ = get_dataloader(__UpperCamelCase ,not dispatch_batches )
A_ = AutoModelForSequenceClassification.from_pretrained(
"hf-internal-testing/mrpc-bert-base-cased" ,return_dict=__UpperCamelCase )
A_ , A_ = accelerator.prepare(__UpperCamelCase ,__UpperCamelCase )
return {"ddp": [ddp_model, ddp_dataloader, "cuda:0"], "no": [model, dataloader, accelerator.device]}, accelerator
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : int ,__UpperCamelCase : Optional[Any] ):
"""simple docstring"""
A_ = []
for batch in dataloader:
A_ , A_ = batch.values()
with torch.no_grad():
A_ = model(__UpperCamelCase )
A_ , A_ = accelerator.gather_for_metrics((logit, target) )
logits_and_targets.append((logit, target) )
A_ , A_ = [], []
for logit, targ in logits_and_targets:
logits.append(__UpperCamelCase )
targs.append(__UpperCamelCase )
A_ , A_ = torch.cat(__UpperCamelCase ), torch.cat(__UpperCamelCase )
return logits, targs
def __snake_case ( __UpperCamelCase : Accelerator ,__UpperCamelCase : Dict=82 ,__UpperCamelCase : List[Any]=False ,__UpperCamelCase : Dict=False ,__UpperCamelCase : Optional[int]=16 ):
"""simple docstring"""
A_ , A_ , A_ = get_basic_setup(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
A_ , A_ = generate_predictions(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
assert (
len(__UpperCamelCase ) == num_samples
), f'''Unexpected number of inputs:\n Expected: {num_samples}\n Actual: {len(__UpperCamelCase )}'''
def __snake_case ( __UpperCamelCase : bool = False ,__UpperCamelCase : bool = False ):
"""simple docstring"""
A_ = evaluate.load("glue" ,"mrpc" )
A_ , A_ = get_mrpc_setup(__UpperCamelCase ,__UpperCamelCase )
# First do baseline
A_ , A_ , A_ = setup["no"]
model.to(__UpperCamelCase )
model.eval()
for batch in dataloader:
batch.to(__UpperCamelCase )
with torch.inference_mode():
A_ = model(**__UpperCamelCase )
A_ = outputs.logits.argmax(dim=-1 )
metric.add_batch(predictions=__UpperCamelCase ,references=batch["labels"] )
A_ = metric.compute()
# Then do distributed
A_ , A_ , A_ = setup["ddp"]
model.eval()
for batch in dataloader:
with torch.inference_mode():
A_ = model(**__UpperCamelCase )
A_ = outputs.logits.argmax(dim=-1 )
A_ = batch["labels"]
A_ , A_ = accelerator.gather_for_metrics((preds, references) )
metric.add_batch(predictions=__UpperCamelCase ,references=__UpperCamelCase )
A_ = metric.compute()
for key in "accuracy f1".split():
assert math.isclose(
baseline[key] ,distributed[key] ), f'''Baseline and Distributed are not the same for key {key}:\n\tBaseline: {baseline[key]}\n\tDistributed: {distributed[key]}\n'''
def __snake_case ( ):
"""simple docstring"""
A_ = Accelerator(split_batches=__UpperCamelCase ,dispatch_batches=__UpperCamelCase )
if accelerator.is_local_main_process:
datasets.utils.logging.set_verbosity_warning()
transformers.utils.logging.set_verbosity_warning()
else:
datasets.utils.logging.set_verbosity_error()
transformers.utils.logging.set_verbosity_error()
# These are a bit slower so they should only be ran on the GPU or TPU
if torch.cuda.is_available() or is_tpu_available():
if accelerator.is_local_main_process:
print("**Testing gather_for_metrics**" )
for split_batches in [True, False]:
for dispatch_batches in [True, False]:
if accelerator.is_local_main_process:
print(f'''With: `split_batches={split_batches}`, `dispatch_batches={dispatch_batches}`''' )
test_mrpc(__UpperCamelCase ,__UpperCamelCase )
accelerator.state._reset_state()
if accelerator.is_local_main_process:
print("**Test torch metrics**" )
for split_batches in [True, False]:
for dispatch_batches in [True, False]:
A_ = Accelerator(split_batches=__UpperCamelCase ,dispatch_batches=__UpperCamelCase )
if accelerator.is_local_main_process:
print(f'''With: `split_batches={split_batches}`, `dispatch_batches={dispatch_batches}`, length=99''' )
test_torch_metrics(__UpperCamelCase ,99 )
accelerator.state._reset_state()
if accelerator.is_local_main_process:
print("**Test last batch is not dropped when perfectly divisible**" )
A_ = Accelerator()
test_torch_metrics(__UpperCamelCase ,512 )
accelerator.state._reset_state()
def __snake_case ( __UpperCamelCase : List[Any] ):
"""simple docstring"""
main()
if __name__ == "__main__":
main() | 86 | 1 |
from bisect import bisect
from itertools import accumulate
def __snake_case ( __UpperCamelCase : int ,__UpperCamelCase : Tuple ,__UpperCamelCase : str ,__UpperCamelCase : str ):
"""simple docstring"""
A_ = sorted(zip(__UpperCamelCase ,__UpperCamelCase ) ,key=lambda __UpperCamelCase : x[0] / x[1] ,reverse=__UpperCamelCase )
A_ , A_ = [i[0] for i in r], [i[1] for i in r]
A_ = list(accumulate(__UpperCamelCase ) )
A_ = bisect(__UpperCamelCase ,__UpperCamelCase )
return (
0
if k == 0
else sum(vl[:k] ) + (w - acc[k - 1]) * (vl[k]) / (wt[k])
if k != n
else sum(vl[:k] )
)
if __name__ == "__main__":
import doctest
doctest.testmod() | 86 |
import argparse
import os
from transformers.utils import direct_transformers_import
# All paths are set with the intent you should run this script from the root of the repo with the command
# python utils/check_task_guides.py
__a :Optional[Any] = 'src/transformers'
__a :Tuple = 'docs/source/en/tasks'
def __snake_case ( __UpperCamelCase : List[Any] ,__UpperCamelCase : Dict ,__UpperCamelCase : int ):
"""simple docstring"""
with open(__UpperCamelCase ,"r" ,encoding="utf-8" ,newline="\n" ) as f:
A_ = f.readlines()
# Find the start prompt.
A_ = 0
while not lines[start_index].startswith(__UpperCamelCase ):
start_index += 1
start_index += 1
A_ = start_index
while not lines[end_index].startswith(__UpperCamelCase ):
end_index += 1
end_index -= 1
while len(lines[start_index] ) <= 1:
start_index += 1
while len(lines[end_index] ) <= 1:
end_index -= 1
end_index += 1
return "".join(lines[start_index:end_index] ), start_index, end_index, lines
# This is to make sure the transformers module imported is the one in the repo.
__a :List[str] = direct_transformers_import(TRANSFORMERS_PATH)
__a :Optional[Any] = {
'asr.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_CTC_MAPPING_NAMES,
'audio_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_AUDIO_CLASSIFICATION_MAPPING_NAMES,
'language_modeling.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_CAUSAL_LM_MAPPING_NAMES,
'image_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_IMAGE_CLASSIFICATION_MAPPING_NAMES,
'masked_language_modeling.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_MASKED_LM_MAPPING_NAMES,
'multiple_choice.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_MULTIPLE_CHOICE_MAPPING_NAMES,
'object_detection.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_OBJECT_DETECTION_MAPPING_NAMES,
'question_answering.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_QUESTION_ANSWERING_MAPPING_NAMES,
'semantic_segmentation.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEMANTIC_SEGMENTATION_MAPPING_NAMES,
'sequence_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING_NAMES,
'summarization.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES,
'token_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING_NAMES,
'translation.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES,
'video_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_VIDEO_CLASSIFICATION_MAPPING_NAMES,
'document_question_answering.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_DOCUMENT_QUESTION_ANSWERING_MAPPING_NAMES,
'monocular_depth_estimation.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_DEPTH_ESTIMATION_MAPPING_NAMES,
}
# This list contains model types used in some task guides that are not in `CONFIG_MAPPING_NAMES` (therefore not in any
# `MODEL_MAPPING_NAMES` or any `MODEL_FOR_XXX_MAPPING_NAMES`).
__a :Optional[Any] = {
'summarization.md': ('nllb',),
'translation.md': ('nllb',),
}
def __snake_case ( __UpperCamelCase : Tuple ):
"""simple docstring"""
A_ = TASK_GUIDE_TO_MODELS[task_guide]
A_ = SPECIAL_TASK_GUIDE_TO_MODEL_TYPES.get(__UpperCamelCase ,set() )
A_ = {
code: name
for code, name in transformers_module.MODEL_NAMES_MAPPING.items()
if (code in model_maping_names or code in special_model_types)
}
return ", ".join([f'''[{name}](../model_doc/{code})''' for code, name in model_names.items()] ) + "\n"
def __snake_case ( __UpperCamelCase : str ,__UpperCamelCase : List[str]=False ):
"""simple docstring"""
A_ , A_ , A_ , A_ = _find_text_in_file(
filename=os.path.join(__UpperCamelCase ,__UpperCamelCase ) ,start_prompt="<!--This tip is automatically generated by `make fix-copies`, do not fill manually!-->" ,end_prompt="<!--End of the generated tip-->" ,)
A_ = get_model_list_for_task(__UpperCamelCase )
if current_list != new_list:
if overwrite:
with open(os.path.join(__UpperCamelCase ,__UpperCamelCase ) ,"w" ,encoding="utf-8" ,newline="\n" ) as f:
f.writelines(lines[:start_index] + [new_list] + lines[end_index:] )
else:
raise ValueError(
f'''The list of models that can be used in the {task_guide} guide needs an update. Run `make fix-copies`'''
" to fix this." )
if __name__ == "__main__":
__a :int = argparse.ArgumentParser()
parser.add_argument('--fix_and_overwrite', action='store_true', help='Whether to fix inconsistencies.')
__a :Optional[Any] = parser.parse_args()
for task_guide in TASK_GUIDE_TO_MODELS.keys():
check_model_list_for_task(task_guide, args.fix_and_overwrite) | 86 | 1 |
import sys
__a :Dict = (
'73167176531330624919225119674426574742355349194934'
'96983520312774506326239578318016984801869478851843'
'85861560789112949495459501737958331952853208805511'
'12540698747158523863050715693290963295227443043557'
'66896648950445244523161731856403098711121722383113'
'62229893423380308135336276614282806444486645238749'
'30358907296290491560440772390713810515859307960866'
'70172427121883998797908792274921901699720888093776'
'65727333001053367881220235421809751254540594752243'
'52584907711670556013604839586446706324415722155397'
'53697817977846174064955149290862569321978468622482'
'83972241375657056057490261407972968652414535100474'
'82166370484403199890008895243450658541227588666881'
'16427171479924442928230863465674813919123162824586'
'17866458359124566529476545682848912883142607690042'
'24219022671055626321111109370544217506941658960408'
'07198403850962455444362981230987879927244284909188'
'84580156166097919133875499200524063689912560717606'
'05886116467109405077541002256983155200055935729725'
'71636269561882670428252483600823257530420752963450'
)
def __snake_case ( __UpperCamelCase : str ):
"""simple docstring"""
A_ = 1
for digit in s:
product *= int(__UpperCamelCase )
return product
def __snake_case ( __UpperCamelCase : str = N ):
"""simple docstring"""
A_ = -sys.maxsize - 1
A_ = n[:13]
A_ = 13
while cur_index < len(__UpperCamelCase ) - 13:
if int(n[cur_index] ) >= int(substr[0] ):
A_ = substr[1:] + n[cur_index]
cur_index += 1
else:
A_ = max(__UpperCamelCase ,str_eval(__UpperCamelCase ) )
A_ = n[cur_index : cur_index + 13]
cur_index += 13
return largest_product
if __name__ == "__main__":
print(F"{solution() = }") | 86 |
import argparse
import json
from pathlib import Path
import requests
import timm
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from transformers import DeiTImageProcessor, ViTConfig, ViTForImageClassification, ViTImageProcessor, ViTModel
from transformers.utils import logging
logging.set_verbosity_info()
__a :Dict = logging.get_logger(__name__)
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : Tuple=False ):
"""simple docstring"""
A_ = []
for i in range(config.num_hidden_layers ):
# encoder layers: output projection, 2 feedforward neural networks and 2 layernorms
rename_keys.append((f'''blocks.{i}.norm1.weight''', f'''vit.encoder.layer.{i}.layernorm_before.weight''') )
rename_keys.append((f'''blocks.{i}.norm1.bias''', f'''vit.encoder.layer.{i}.layernorm_before.bias''') )
rename_keys.append((f'''blocks.{i}.attn.proj.weight''', f'''vit.encoder.layer.{i}.attention.output.dense.weight''') )
rename_keys.append((f'''blocks.{i}.attn.proj.bias''', f'''vit.encoder.layer.{i}.attention.output.dense.bias''') )
rename_keys.append((f'''blocks.{i}.norm2.weight''', f'''vit.encoder.layer.{i}.layernorm_after.weight''') )
rename_keys.append((f'''blocks.{i}.norm2.bias''', f'''vit.encoder.layer.{i}.layernorm_after.bias''') )
rename_keys.append((f'''blocks.{i}.mlp.fc1.weight''', f'''vit.encoder.layer.{i}.intermediate.dense.weight''') )
rename_keys.append((f'''blocks.{i}.mlp.fc1.bias''', f'''vit.encoder.layer.{i}.intermediate.dense.bias''') )
rename_keys.append((f'''blocks.{i}.mlp.fc2.weight''', f'''vit.encoder.layer.{i}.output.dense.weight''') )
rename_keys.append((f'''blocks.{i}.mlp.fc2.bias''', f'''vit.encoder.layer.{i}.output.dense.bias''') )
# projection layer + position embeddings
rename_keys.extend(
[
("cls_token", "vit.embeddings.cls_token"),
("patch_embed.proj.weight", "vit.embeddings.patch_embeddings.projection.weight"),
("patch_embed.proj.bias", "vit.embeddings.patch_embeddings.projection.bias"),
("pos_embed", "vit.embeddings.position_embeddings"),
] )
if base_model:
# layernorm + pooler
rename_keys.extend(
[
("norm.weight", "layernorm.weight"),
("norm.bias", "layernorm.bias"),
("pre_logits.fc.weight", "pooler.dense.weight"),
("pre_logits.fc.bias", "pooler.dense.bias"),
] )
# if just the base model, we should remove "vit" from all keys that start with "vit"
A_ = [(pair[0], pair[1][4:]) if pair[1].startswith("vit" ) else pair for pair in rename_keys]
else:
# layernorm + classification head
rename_keys.extend(
[
("norm.weight", "vit.layernorm.weight"),
("norm.bias", "vit.layernorm.bias"),
("head.weight", "classifier.weight"),
("head.bias", "classifier.bias"),
] )
return rename_keys
def __snake_case ( __UpperCamelCase : Optional[int] ,__UpperCamelCase : Tuple ,__UpperCamelCase : Any=False ):
"""simple docstring"""
for i in range(config.num_hidden_layers ):
if base_model:
A_ = ""
else:
A_ = "vit."
# read in weights + bias of input projection layer (in timm, this is a single matrix + bias)
A_ = state_dict.pop(f'''blocks.{i}.attn.qkv.weight''' )
A_ = state_dict.pop(f'''blocks.{i}.attn.qkv.bias''' )
# next, add query, keys and values (in that order) to the state dict
A_ = in_proj_weight[
: config.hidden_size, :
]
A_ = in_proj_bias[: config.hidden_size]
A_ = in_proj_weight[
config.hidden_size : config.hidden_size * 2, :
]
A_ = in_proj_bias[
config.hidden_size : config.hidden_size * 2
]
A_ = in_proj_weight[
-config.hidden_size :, :
]
A_ = in_proj_bias[-config.hidden_size :]
def __snake_case ( __UpperCamelCase : List[Any] ):
"""simple docstring"""
A_ = ["head.weight", "head.bias"]
for k in ignore_keys:
state_dict.pop(__UpperCamelCase ,__UpperCamelCase )
def __snake_case ( __UpperCamelCase : Any ,__UpperCamelCase : Optional[int] ,__UpperCamelCase : List[str] ):
"""simple docstring"""
A_ = dct.pop(__UpperCamelCase )
A_ = val
def __snake_case ( ):
"""simple docstring"""
A_ = "http://images.cocodataset.org/val2017/000000039769.jpg"
A_ = Image.open(requests.get(__UpperCamelCase ,stream=__UpperCamelCase ).raw )
return im
@torch.no_grad()
def __snake_case ( __UpperCamelCase : Union[str, Any] ,__UpperCamelCase : Optional[int] ):
"""simple docstring"""
A_ = ViTConfig()
A_ = False
# dataset (ImageNet-21k only or also fine-tuned on ImageNet 2012), patch_size and image_size
if vit_name[-5:] == "in21k":
A_ = True
A_ = int(vit_name[-12:-10] )
A_ = int(vit_name[-9:-6] )
else:
A_ = 1000
A_ = "huggingface/label-files"
A_ = "imagenet-1k-id2label.json"
A_ = json.load(open(hf_hub_download(__UpperCamelCase ,__UpperCamelCase ,repo_type="dataset" ) ,"r" ) )
A_ = {int(__UpperCamelCase ): v for k, v in idalabel.items()}
A_ = idalabel
A_ = {v: k for k, v in idalabel.items()}
A_ = int(vit_name[-6:-4] )
A_ = int(vit_name[-3:] )
# size of the architecture
if "deit" in vit_name:
if vit_name[9:].startswith("tiny" ):
A_ = 192
A_ = 768
A_ = 12
A_ = 3
elif vit_name[9:].startswith("small" ):
A_ = 384
A_ = 1536
A_ = 12
A_ = 6
else:
pass
else:
if vit_name[4:].startswith("small" ):
A_ = 768
A_ = 2304
A_ = 8
A_ = 8
elif vit_name[4:].startswith("base" ):
pass
elif vit_name[4:].startswith("large" ):
A_ = 1024
A_ = 4096
A_ = 24
A_ = 16
elif vit_name[4:].startswith("huge" ):
A_ = 1280
A_ = 5120
A_ = 32
A_ = 16
# load original model from timm
A_ = timm.create_model(__UpperCamelCase ,pretrained=__UpperCamelCase )
timm_model.eval()
# load state_dict of original model, remove and rename some keys
A_ = timm_model.state_dict()
if base_model:
remove_classification_head_(__UpperCamelCase )
A_ = create_rename_keys(__UpperCamelCase ,__UpperCamelCase )
for src, dest in rename_keys:
rename_key(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
read_in_q_k_v(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
# load HuggingFace model
if vit_name[-5:] == "in21k":
A_ = ViTModel(__UpperCamelCase ).eval()
else:
A_ = ViTForImageClassification(__UpperCamelCase ).eval()
model.load_state_dict(__UpperCamelCase )
# Check outputs on an image, prepared by ViTImageProcessor/DeiTImageProcessor
if "deit" in vit_name:
A_ = DeiTImageProcessor(size=config.image_size )
else:
A_ = ViTImageProcessor(size=config.image_size )
A_ = image_processor(images=prepare_img() ,return_tensors="pt" )
A_ = encoding["pixel_values"]
A_ = model(__UpperCamelCase )
if base_model:
A_ = timm_model.forward_features(__UpperCamelCase )
assert timm_pooled_output.shape == outputs.pooler_output.shape
assert torch.allclose(__UpperCamelCase ,outputs.pooler_output ,atol=1E-3 )
else:
A_ = timm_model(__UpperCamelCase )
assert timm_logits.shape == outputs.logits.shape
assert torch.allclose(__UpperCamelCase ,outputs.logits ,atol=1E-3 )
Path(__UpperCamelCase ).mkdir(exist_ok=__UpperCamelCase )
print(f'''Saving model {vit_name} to {pytorch_dump_folder_path}''' )
model.save_pretrained(__UpperCamelCase )
print(f'''Saving image processor to {pytorch_dump_folder_path}''' )
image_processor.save_pretrained(__UpperCamelCase )
if __name__ == "__main__":
__a :str = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
'--vit_name',
default='vit_base_patch16_224',
type=str,
help='Name of the ViT timm model you\'d like to convert.',
)
parser.add_argument(
'--pytorch_dump_folder_path', default=None, type=str, help='Path to the output PyTorch model directory.'
)
__a :Optional[int] = parser.parse_args()
convert_vit_checkpoint(args.vit_name, args.pytorch_dump_folder_path) | 86 | 1 |
import json
import os
import shutil
import tempfile
import unittest
import numpy as np
import pytest
from transformers import CLIPTokenizer, CLIPTokenizerFast
from transformers.models.clip.tokenization_clip import VOCAB_FILES_NAMES
from transformers.testing_utils import require_vision
from transformers.utils import IMAGE_PROCESSOR_NAME, is_vision_available
if is_vision_available():
from PIL import Image
from transformers import OwlViTImageProcessor, OwlViTProcessor
@require_vision
class _a ( unittest.TestCase ):
"""simple docstring"""
def __A ( self : List[Any] ):
A_ = tempfile.mkdtemp()
# fmt: off
A_ = ["", "l", "o", "w", "e", "r", "s", "t", "i", "d", "n", "lo", "l</w>", "w</w>", "r</w>", "t</w>", "low</w>", "er</w>", "lowest</w>", "newer</w>", "wider", "<unk>", "<|startoftext|>", "<|endoftext|>"]
# fmt: on
A_ = dict(zip(UpperCAmelCase , range(len(UpperCAmelCase ) ) ) )
A_ = ["#version: 0.2", "l o", "lo w</w>", "e r</w>", ""]
A_ = {"unk_token": "<unk>"}
A_ = os.path.join(self.tmpdirname , VOCAB_FILES_NAMES["vocab_file"] )
A_ = os.path.join(self.tmpdirname , VOCAB_FILES_NAMES["merges_file"] )
with open(self.vocab_file , "w" , encoding="utf-8" ) as fp:
fp.write(json.dumps(UpperCAmelCase ) + "\n" )
with open(self.merges_file , "w" , encoding="utf-8" ) as fp:
fp.write("\n".join(UpperCAmelCase ) )
A_ = {
"do_resize": True,
"size": 20,
"do_center_crop": True,
"crop_size": 18,
"do_normalize": True,
"image_mean": [0.48_145_466, 0.4_578_275, 0.40_821_073],
"image_std": [0.26_862_954, 0.26_130_258, 0.27_577_711],
}
A_ = os.path.join(self.tmpdirname , UpperCAmelCase )
with open(self.image_processor_file , "w" , encoding="utf-8" ) as fp:
json.dump(UpperCAmelCase , UpperCAmelCase )
def __A ( self : Any , **UpperCAmelCase : str ):
return CLIPTokenizer.from_pretrained(self.tmpdirname , pad_token="!" , **UpperCAmelCase )
def __A ( self : Dict , **UpperCAmelCase : Optional[Any] ):
return CLIPTokenizerFast.from_pretrained(self.tmpdirname , pad_token="!" , **UpperCAmelCase )
def __A ( self : List[Any] , **UpperCAmelCase : str ):
return OwlViTImageProcessor.from_pretrained(self.tmpdirname , **UpperCAmelCase )
def __A ( self : List[Any] ):
shutil.rmtree(self.tmpdirname )
def __A ( self : Tuple ):
A_ = [np.random.randint(255 , size=(3, 30, 400) , dtype=np.uinta )]
A_ = [Image.fromarray(np.moveaxis(UpperCAmelCase , 0 , -1 ) ) for x in image_inputs]
return image_inputs
def __A ( self : Any ):
A_ = self.get_tokenizer()
A_ = self.get_rust_tokenizer()
A_ = self.get_image_processor()
A_ = OwlViTProcessor(tokenizer=UpperCAmelCase , image_processor=UpperCAmelCase )
processor_slow.save_pretrained(self.tmpdirname )
A_ = OwlViTProcessor.from_pretrained(self.tmpdirname , use_fast=UpperCAmelCase )
A_ = OwlViTProcessor(tokenizer=UpperCAmelCase , image_processor=UpperCAmelCase )
processor_fast.save_pretrained(self.tmpdirname )
A_ = OwlViTProcessor.from_pretrained(self.tmpdirname )
self.assertEqual(processor_slow.tokenizer.get_vocab() , tokenizer_slow.get_vocab() )
self.assertEqual(processor_fast.tokenizer.get_vocab() , tokenizer_fast.get_vocab() )
self.assertEqual(tokenizer_slow.get_vocab() , tokenizer_fast.get_vocab() )
self.assertIsInstance(processor_slow.tokenizer , UpperCAmelCase )
self.assertIsInstance(processor_fast.tokenizer , UpperCAmelCase )
self.assertEqual(processor_slow.image_processor.to_json_string() , image_processor.to_json_string() )
self.assertEqual(processor_fast.image_processor.to_json_string() , image_processor.to_json_string() )
self.assertIsInstance(processor_slow.image_processor , UpperCAmelCase )
self.assertIsInstance(processor_fast.image_processor , UpperCAmelCase )
def __A ( self : Any ):
A_ = OwlViTProcessor(tokenizer=self.get_tokenizer() , image_processor=self.get_image_processor() )
processor.save_pretrained(self.tmpdirname )
A_ = self.get_tokenizer(bos_token="(BOS)" , eos_token="(EOS)" )
A_ = self.get_image_processor(do_normalize=UpperCAmelCase )
A_ = OwlViTProcessor.from_pretrained(
self.tmpdirname , bos_token="(BOS)" , eos_token="(EOS)" , do_normalize=UpperCAmelCase )
self.assertEqual(processor.tokenizer.get_vocab() , tokenizer_add_kwargs.get_vocab() )
self.assertIsInstance(processor.tokenizer , UpperCAmelCase )
self.assertEqual(processor.image_processor.to_json_string() , image_processor_add_kwargs.to_json_string() )
self.assertIsInstance(processor.image_processor , UpperCAmelCase )
def __A ( self : Any ):
A_ = self.get_image_processor()
A_ = self.get_tokenizer()
A_ = OwlViTProcessor(tokenizer=UpperCAmelCase , image_processor=UpperCAmelCase )
A_ = self.prepare_image_inputs()
A_ = image_processor(UpperCAmelCase , return_tensors="np" )
A_ = processor(images=UpperCAmelCase , return_tensors="np" )
for key in input_image_proc.keys():
self.assertAlmostEqual(input_image_proc[key].sum() , input_processor[key].sum() , delta=1E-2 )
def __A ( self : Optional[int] ):
A_ = self.get_image_processor()
A_ = self.get_tokenizer()
A_ = OwlViTProcessor(tokenizer=UpperCAmelCase , image_processor=UpperCAmelCase )
A_ = "lower newer"
A_ = processor(text=UpperCAmelCase , return_tensors="np" )
A_ = tokenizer(UpperCAmelCase , return_tensors="np" )
for key in encoded_tok.keys():
self.assertListEqual(encoded_tok[key][0].tolist() , encoded_processor[key][0].tolist() )
def __A ( self : int ):
A_ = self.get_image_processor()
A_ = self.get_tokenizer()
A_ = OwlViTProcessor(tokenizer=UpperCAmelCase , image_processor=UpperCAmelCase )
A_ = "lower newer"
A_ = self.prepare_image_inputs()
A_ = processor(text=UpperCAmelCase , images=UpperCAmelCase )
self.assertListEqual(list(inputs.keys() ) , ["input_ids", "attention_mask", "pixel_values"] )
# test if it raises when no input is passed
with pytest.raises(UpperCAmelCase ):
processor()
def __A ( self : int ):
A_ = "google/owlvit-base-patch32"
A_ = OwlViTProcessor.from_pretrained(UpperCAmelCase )
A_ = ["cat", "nasa badge"]
A_ = processor(text=UpperCAmelCase )
A_ = 16
self.assertListEqual(list(inputs.keys() ) , ["input_ids", "attention_mask"] )
self.assertEqual(inputs["input_ids"].shape , (2, seq_length) )
# test if it raises when no input is passed
with pytest.raises(UpperCAmelCase ):
processor()
def __A ( self : Optional[int] ):
A_ = "google/owlvit-base-patch32"
A_ = OwlViTProcessor.from_pretrained(UpperCAmelCase )
A_ = [["cat", "nasa badge"], ["person"]]
A_ = processor(text=UpperCAmelCase )
A_ = 16
A_ = len(UpperCAmelCase )
A_ = max([len(UpperCAmelCase ) for texts in input_texts] )
self.assertListEqual(list(inputs.keys() ) , ["input_ids", "attention_mask"] )
self.assertEqual(inputs["input_ids"].shape , (batch_size * num_max_text_queries, seq_length) )
# test if it raises when no input is passed
with pytest.raises(UpperCAmelCase ):
processor()
def __A ( self : Optional[Any] ):
A_ = "google/owlvit-base-patch32"
A_ = OwlViTProcessor.from_pretrained(UpperCAmelCase )
A_ = ["cat", "nasa badge"]
A_ = processor(text=UpperCAmelCase )
A_ = 16
A_ = inputs["input_ids"]
A_ = [
[49406, 2368, 49407, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[49406, 6841, 11301, 49407, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
]
self.assertListEqual(list(inputs.keys() ) , ["input_ids", "attention_mask"] )
self.assertEqual(inputs["input_ids"].shape , (2, seq_length) )
self.assertListEqual(list(input_ids[0] ) , predicted_ids[0] )
self.assertListEqual(list(input_ids[1] ) , predicted_ids[1] )
def __A ( self : Tuple ):
A_ = self.get_image_processor()
A_ = self.get_tokenizer()
A_ = OwlViTProcessor(tokenizer=UpperCAmelCase , image_processor=UpperCAmelCase )
A_ = self.prepare_image_inputs()
A_ = self.prepare_image_inputs()
A_ = processor(images=UpperCAmelCase , query_images=UpperCAmelCase )
self.assertListEqual(list(inputs.keys() ) , ["query_pixel_values", "pixel_values"] )
# test if it raises when no input is passed
with pytest.raises(UpperCAmelCase ):
processor()
def __A ( self : List[str] ):
A_ = self.get_image_processor()
A_ = self.get_tokenizer()
A_ = OwlViTProcessor(tokenizer=UpperCAmelCase , image_processor=UpperCAmelCase )
A_ = [[1, 4, 5, 8, 1, 0, 8], [3, 4, 3, 1, 1, 8, 9]]
A_ = processor.batch_decode(UpperCAmelCase )
A_ = tokenizer.batch_decode(UpperCAmelCase )
self.assertListEqual(UpperCAmelCase , UpperCAmelCase ) | 86 |
def __snake_case ( __UpperCamelCase : int = 50 ):
"""simple docstring"""
A_ = [1] * (length + 1)
for row_length in range(length + 1 ):
for tile_length in range(2 ,5 ):
for tile_start in range(row_length - tile_length + 1 ):
ways_number[row_length] += ways_number[
row_length - tile_start - tile_length
]
return ways_number[length]
if __name__ == "__main__":
print(F"{solution() = }") | 86 | 1 |
from typing import Optional, Tuple, Union
import flax
import flax.linen as nn
import jax
import jax.numpy as jnp
from flax.core.frozen_dict import FrozenDict
from ..configuration_utils import ConfigMixin, flax_register_to_config
from ..utils import BaseOutput
from .embeddings_flax import FlaxTimestepEmbedding, FlaxTimesteps
from .modeling_flax_utils import FlaxModelMixin
from .unet_ad_blocks_flax import (
FlaxCrossAttnDownBlockaD,
FlaxDownBlockaD,
FlaxUNetMidBlockaDCrossAttn,
)
@flax.struct.dataclass
class _a ( snake_case_ ):
"""simple docstring"""
_lowerCamelCase : jnp.ndarray
_lowerCamelCase : jnp.ndarray
class _a ( nn.Module ):
"""simple docstring"""
_lowerCamelCase : int
_lowerCamelCase : Tuple[int] = (1_6, 3_2, 9_6, 2_5_6)
_lowerCamelCase : jnp.dtype = jnp.floataa
def __A ( self : Optional[Any] ):
A_ = nn.Conv(
self.block_out_channels[0] , kernel_size=(3, 3) , padding=((1, 1), (1, 1)) , dtype=self.dtype , )
A_ = []
for i in range(len(self.block_out_channels ) - 1 ):
A_ = self.block_out_channels[i]
A_ = self.block_out_channels[i + 1]
A_ = nn.Conv(
UpperCAmelCase , kernel_size=(3, 3) , padding=((1, 1), (1, 1)) , dtype=self.dtype , )
blocks.append(UpperCAmelCase )
A_ = nn.Conv(
UpperCAmelCase , kernel_size=(3, 3) , strides=(2, 2) , padding=((1, 1), (1, 1)) , dtype=self.dtype , )
blocks.append(UpperCAmelCase )
A_ = blocks
A_ = nn.Conv(
self.conditioning_embedding_channels , kernel_size=(3, 3) , padding=((1, 1), (1, 1)) , kernel_init=nn.initializers.zeros_init() , bias_init=nn.initializers.zeros_init() , dtype=self.dtype , )
def __call__( self : int , UpperCAmelCase : int ):
A_ = self.conv_in(UpperCAmelCase )
A_ = nn.silu(UpperCAmelCase )
for block in self.blocks:
A_ = block(UpperCAmelCase )
A_ = nn.silu(UpperCAmelCase )
A_ = self.conv_out(UpperCAmelCase )
return embedding
@flax_register_to_config
class _a ( nn.Module , snake_case_ , snake_case_ ):
"""simple docstring"""
_lowerCamelCase : int = 3_2
_lowerCamelCase : int = 4
_lowerCamelCase : Tuple[str] = (
"CrossAttnDownBlock2D",
"CrossAttnDownBlock2D",
"CrossAttnDownBlock2D",
"DownBlock2D",
)
_lowerCamelCase : Union[bool, Tuple[bool]] = False
_lowerCamelCase : Tuple[int] = (3_2_0, 6_4_0, 1_2_8_0, 1_2_8_0)
_lowerCamelCase : int = 2
_lowerCamelCase : Union[int, Tuple[int]] = 8
_lowerCamelCase : Optional[Union[int, Tuple[int]]] = None
_lowerCamelCase : int = 1_2_8_0
_lowerCamelCase : float = 0.0
_lowerCamelCase : bool = False
_lowerCamelCase : jnp.dtype = jnp.floataa
_lowerCamelCase : bool = True
_lowerCamelCase : int = 0
_lowerCamelCase : str = "rgb"
_lowerCamelCase : Tuple[int] = (1_6, 3_2, 9_6, 2_5_6)
def __A ( self : List[Any] , UpperCAmelCase : jax.random.KeyArray ):
# init input tensors
A_ = (1, self.in_channels, self.sample_size, self.sample_size)
A_ = jnp.zeros(UpperCAmelCase , dtype=jnp.floataa )
A_ = jnp.ones((1,) , dtype=jnp.intaa )
A_ = jnp.zeros((1, 1, self.cross_attention_dim) , dtype=jnp.floataa )
A_ = (1, 3, self.sample_size * 8, self.sample_size * 8)
A_ = jnp.zeros(UpperCAmelCase , dtype=jnp.floataa )
A_ , A_ = jax.random.split(UpperCAmelCase )
A_ = {"params": params_rng, "dropout": dropout_rng}
return self.init(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase , UpperCAmelCase , UpperCAmelCase )["params"]
def __A ( self : str ):
A_ = self.block_out_channels
A_ = block_out_channels[0] * 4
# If `num_attention_heads` is not defined (which is the case for most models)
# it will default to `attention_head_dim`. This looks weird upon first reading it and it is.
# The reason for this behavior is to correct for incorrectly named variables that were introduced
# when this library was created. The incorrect naming was only discovered much later in https://github.com/huggingface/diffusers/issues/2011#issuecomment-1547958131
# Changing `attention_head_dim` to `num_attention_heads` for 40,000+ configurations is too backwards breaking
# which is why we correct for the naming here.
A_ = self.num_attention_heads or self.attention_head_dim
# input
A_ = nn.Conv(
block_out_channels[0] , kernel_size=(3, 3) , strides=(1, 1) , padding=((1, 1), (1, 1)) , dtype=self.dtype , )
# time
A_ = FlaxTimesteps(
block_out_channels[0] , flip_sin_to_cos=self.flip_sin_to_cos , freq_shift=self.config.freq_shift )
A_ = FlaxTimestepEmbedding(UpperCAmelCase , dtype=self.dtype )
A_ = FlaxControlNetConditioningEmbedding(
conditioning_embedding_channels=block_out_channels[0] , block_out_channels=self.conditioning_embedding_out_channels , )
A_ = self.only_cross_attention
if isinstance(UpperCAmelCase , UpperCAmelCase ):
A_ = (only_cross_attention,) * len(self.down_block_types )
if isinstance(UpperCAmelCase , UpperCAmelCase ):
A_ = (num_attention_heads,) * len(self.down_block_types )
# down
A_ = []
A_ = []
A_ = block_out_channels[0]
A_ = nn.Conv(
UpperCAmelCase , kernel_size=(1, 1) , padding="VALID" , kernel_init=nn.initializers.zeros_init() , bias_init=nn.initializers.zeros_init() , dtype=self.dtype , )
controlnet_down_blocks.append(UpperCAmelCase )
for i, down_block_type in enumerate(self.down_block_types ):
A_ = output_channel
A_ = block_out_channels[i]
A_ = i == len(UpperCAmelCase ) - 1
if down_block_type == "CrossAttnDownBlock2D":
A_ = FlaxCrossAttnDownBlockaD(
in_channels=UpperCAmelCase , out_channels=UpperCAmelCase , dropout=self.dropout , num_layers=self.layers_per_block , num_attention_heads=num_attention_heads[i] , add_downsample=not is_final_block , use_linear_projection=self.use_linear_projection , only_cross_attention=only_cross_attention[i] , dtype=self.dtype , )
else:
A_ = FlaxDownBlockaD(
in_channels=UpperCAmelCase , out_channels=UpperCAmelCase , dropout=self.dropout , num_layers=self.layers_per_block , add_downsample=not is_final_block , dtype=self.dtype , )
down_blocks.append(UpperCAmelCase )
for _ in range(self.layers_per_block ):
A_ = nn.Conv(
UpperCAmelCase , kernel_size=(1, 1) , padding="VALID" , kernel_init=nn.initializers.zeros_init() , bias_init=nn.initializers.zeros_init() , dtype=self.dtype , )
controlnet_down_blocks.append(UpperCAmelCase )
if not is_final_block:
A_ = nn.Conv(
UpperCAmelCase , kernel_size=(1, 1) , padding="VALID" , kernel_init=nn.initializers.zeros_init() , bias_init=nn.initializers.zeros_init() , dtype=self.dtype , )
controlnet_down_blocks.append(UpperCAmelCase )
A_ = down_blocks
A_ = controlnet_down_blocks
# mid
A_ = block_out_channels[-1]
A_ = FlaxUNetMidBlockaDCrossAttn(
in_channels=UpperCAmelCase , dropout=self.dropout , num_attention_heads=num_attention_heads[-1] , use_linear_projection=self.use_linear_projection , dtype=self.dtype , )
A_ = nn.Conv(
UpperCAmelCase , kernel_size=(1, 1) , padding="VALID" , kernel_init=nn.initializers.zeros_init() , bias_init=nn.initializers.zeros_init() , dtype=self.dtype , )
def __call__( self : str , UpperCAmelCase : Optional[int] , UpperCAmelCase : Union[str, Any] , UpperCAmelCase : Optional[int] , UpperCAmelCase : Dict , UpperCAmelCase : float = 1.0 , UpperCAmelCase : bool = True , UpperCAmelCase : bool = False , ):
A_ = self.controlnet_conditioning_channel_order
if channel_order == "bgr":
A_ = jnp.flip(UpperCAmelCase , axis=1 )
# 1. time
if not isinstance(UpperCAmelCase , jnp.ndarray ):
A_ = jnp.array([timesteps] , dtype=jnp.intaa )
elif isinstance(UpperCAmelCase , jnp.ndarray ) and len(timesteps.shape ) == 0:
A_ = timesteps.astype(dtype=jnp.floataa )
A_ = jnp.expand_dims(UpperCAmelCase , 0 )
A_ = self.time_proj(UpperCAmelCase )
A_ = self.time_embedding(UpperCAmelCase )
# 2. pre-process
A_ = jnp.transpose(UpperCAmelCase , (0, 2, 3, 1) )
A_ = self.conv_in(UpperCAmelCase )
A_ = jnp.transpose(UpperCAmelCase , (0, 2, 3, 1) )
A_ = self.controlnet_cond_embedding(UpperCAmelCase )
sample += controlnet_cond
# 3. down
A_ = (sample,)
for down_block in self.down_blocks:
if isinstance(UpperCAmelCase , UpperCAmelCase ):
A_ , A_ = down_block(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase , deterministic=not train )
else:
A_ , A_ = down_block(UpperCAmelCase , UpperCAmelCase , deterministic=not train )
down_block_res_samples += res_samples
# 4. mid
A_ = self.mid_block(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase , deterministic=not train )
# 5. contronet blocks
A_ = ()
for down_block_res_sample, controlnet_block in zip(UpperCAmelCase , self.controlnet_down_blocks ):
A_ = controlnet_block(UpperCAmelCase )
controlnet_down_block_res_samples += (down_block_res_sample,)
A_ = controlnet_down_block_res_samples
A_ = self.controlnet_mid_block(UpperCAmelCase )
# 6. scaling
A_ = [sample * conditioning_scale for sample in down_block_res_samples]
mid_block_res_sample *= conditioning_scale
if not return_dict:
return (down_block_res_samples, mid_block_res_sample)
return FlaxControlNetOutput(
down_block_res_samples=UpperCAmelCase , mid_block_res_sample=UpperCAmelCase ) | 86 |
from typing import Any, Dict, List, Union
from ..utils import add_end_docstrings, is_torch_available, is_vision_available, logging, requires_backends
from .base import PIPELINE_INIT_ARGS, ChunkPipeline
if is_vision_available():
from PIL import Image
from ..image_utils import load_image
if is_torch_available():
import torch
from transformers.modeling_outputs import BaseModelOutput
from ..models.auto.modeling_auto import MODEL_FOR_ZERO_SHOT_OBJECT_DETECTION_MAPPING
__a :List[str] = logging.get_logger(__name__)
@add_end_docstrings(snake_case_ )
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Any , **UpperCAmelCase : List[str] ):
super().__init__(**UpperCAmelCase )
if self.framework == "tf":
raise ValueError(f'''The {self.__class__} is only available in PyTorch.''' )
requires_backends(self , "vision" )
self.check_model_type(UpperCAmelCase )
def __call__( self : Optional[int] , UpperCAmelCase : Union[str, "Image.Image", List[Dict[str, Any]]] , UpperCAmelCase : Union[str, List[str]] = None , **UpperCAmelCase : List[Any] , ):
if "text_queries" in kwargs:
A_ = kwargs.pop("text_queries" )
if isinstance(UpperCAmelCase , (str, Image.Image) ):
A_ = {"image": image, "candidate_labels": candidate_labels}
else:
A_ = image
A_ = super().__call__(UpperCAmelCase , **UpperCAmelCase )
return results
def __A ( self : int , **UpperCAmelCase : Tuple ):
A_ = {}
if "threshold" in kwargs:
A_ = kwargs["threshold"]
if "top_k" in kwargs:
A_ = kwargs["top_k"]
return {}, {}, postprocess_params
def __A ( self : List[str] , UpperCAmelCase : Dict ):
A_ = load_image(inputs["image"] )
A_ = inputs["candidate_labels"]
if isinstance(UpperCAmelCase , UpperCAmelCase ):
A_ = candidate_labels.split("," )
A_ = torch.tensor([[image.height, image.width]] , dtype=torch.intaa )
for i, candidate_label in enumerate(UpperCAmelCase ):
A_ = self.tokenizer(UpperCAmelCase , return_tensors=self.framework )
A_ = self.image_processor(UpperCAmelCase , return_tensors=self.framework )
yield {
"is_last": i == len(UpperCAmelCase ) - 1,
"target_size": target_size,
"candidate_label": candidate_label,
**text_inputs,
**image_features,
}
def __A ( self : str , UpperCAmelCase : int ):
A_ = model_inputs.pop("target_size" )
A_ = model_inputs.pop("candidate_label" )
A_ = model_inputs.pop("is_last" )
A_ = self.model(**UpperCAmelCase )
A_ = {"target_size": target_size, "candidate_label": candidate_label, "is_last": is_last, **outputs}
return model_outputs
def __A ( self : Dict , UpperCAmelCase : Any , UpperCAmelCase : Optional[int]=0.1 , UpperCAmelCase : Optional[int]=None ):
A_ = []
for model_output in model_outputs:
A_ = model_output["candidate_label"]
A_ = BaseModelOutput(UpperCAmelCase )
A_ = self.image_processor.post_process_object_detection(
outputs=UpperCAmelCase , threshold=UpperCAmelCase , target_sizes=model_output["target_size"] )[0]
for index in outputs["scores"].nonzero():
A_ = outputs["scores"][index].item()
A_ = self._get_bounding_box(outputs["boxes"][index][0] )
A_ = {"score": score, "label": label, "box": box}
results.append(UpperCAmelCase )
A_ = sorted(UpperCAmelCase , key=lambda UpperCAmelCase : x["score"] , reverse=UpperCAmelCase )
if top_k:
A_ = results[:top_k]
return results
def __A ( self : List[str] , UpperCAmelCase : "torch.Tensor" ):
if self.framework != "pt":
raise ValueError("The ZeroShotObjectDetectionPipeline is only available in PyTorch." )
A_ , A_ , A_ , A_ = box.int().tolist()
A_ = {
"xmin": xmin,
"ymin": ymin,
"xmax": xmax,
"ymax": ymax,
}
return bbox | 86 | 1 |
def __snake_case ( __UpperCamelCase : int = 100 ):
"""simple docstring"""
A_ = n * (n + 1) * (2 * n + 1) / 6
A_ = (n * (n + 1) / 2) ** 2
return int(square_of_sum - sum_of_squares )
if __name__ == "__main__":
print(F"{solution() = }") | 86 |
import argparse
import torch
# Step 1. clone https://github.com/microsoft/unilm
# Step 2. git checkout to https://github.com/microsoft/unilm/commit/b94ec76c36f02fb2b0bf0dcb0b8554a2185173cd
# Step 3. cd unilm
# Step 4. ln -s $(realpath wavlm/modules.py) ./ # create simlink
# import classes
from unilm.wavlm.WavLM import WavLM as WavLMOrig
from unilm.wavlm.WavLM import WavLMConfig as WavLMConfigOrig
from transformers import WavLMConfig, WavLMModel, logging
logging.set_verbosity_info()
__a :Any = logging.get_logger(__name__)
__a :int = {
'post_extract_proj': 'feature_projection.projection',
'encoder.pos_conv.0': 'encoder.pos_conv_embed.conv',
'self_attn.k_proj': 'encoder.layers.*.attention.k_proj',
'self_attn.v_proj': 'encoder.layers.*.attention.v_proj',
'self_attn.q_proj': 'encoder.layers.*.attention.q_proj',
'self_attn.out_proj': 'encoder.layers.*.attention.out_proj',
'self_attn.grep_linear': 'encoder.layers.*.attention.gru_rel_pos_linear',
'self_attn.relative_attention_bias': 'encoder.layers.*.attention.rel_attn_embed',
'self_attn.grep_a': 'encoder.layers.*.attention.gru_rel_pos_const',
'self_attn_layer_norm': 'encoder.layers.*.layer_norm',
'fc1': 'encoder.layers.*.feed_forward.intermediate_dense',
'fc2': 'encoder.layers.*.feed_forward.output_dense',
'final_layer_norm': 'encoder.layers.*.final_layer_norm',
'encoder.layer_norm': 'encoder.layer_norm',
'w2v_model.layer_norm': 'feature_projection.layer_norm',
'quantizer.weight_proj': 'quantizer.weight_proj',
'quantizer.vars': 'quantizer.codevectors',
'project_q': 'project_q',
'final_proj': 'project_hid',
'w2v_encoder.proj': 'ctc_proj',
'mask_emb': 'masked_spec_embed',
}
__a :Tuple = [
'ctc_proj',
'quantizer.weight_proj',
'quantizer.codevectors',
'project_q',
'project_hid',
]
def __snake_case ( __UpperCamelCase : List[str] ,__UpperCamelCase : List[Any] ,__UpperCamelCase : Dict ,__UpperCamelCase : Any ,__UpperCamelCase : Optional[int] ):
"""simple docstring"""
for attribute in key.split("." ):
A_ = getattr(__UpperCamelCase ,__UpperCamelCase )
if weight_type is not None:
A_ = getattr(__UpperCamelCase ,__UpperCamelCase ).shape
else:
A_ = hf_pointer.shape
assert hf_shape == value.shape, (
f'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
f''' {value.shape} for {full_name}'''
)
if weight_type == "weight":
A_ = value
elif weight_type == "weight_g":
A_ = value
elif weight_type == "weight_v":
A_ = value
elif weight_type == "bias":
A_ = value
else:
A_ = value
logger.info(f'''{key + "." + weight_type if weight_type is not None else ""} was initialized from {full_name}.''' )
def __snake_case ( __UpperCamelCase : Optional[int] ,__UpperCamelCase : Optional[Any] ):
"""simple docstring"""
A_ = []
A_ = fairseq_model.state_dict()
A_ = hf_model.feature_extractor
for name, value in fairseq_dict.items():
A_ = False
if "conv_layers" in name:
load_conv_layer(
__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,hf_model.config.feat_extract_norm == "group" ,)
A_ = True
else:
for key, mapped_key in MAPPING.items():
if key in name or key.split("w2v_model." )[-1] == name.split("." )[0]:
A_ = True
if "*" in mapped_key:
A_ = name.split(__UpperCamelCase )[0].split("." )[-2]
A_ = mapped_key.replace("*" ,__UpperCamelCase )
if "weight_g" in name:
A_ = "weight_g"
elif "weight_v" in name:
A_ = "weight_v"
elif "bias" in name and "relative_attention_bias" not in name:
A_ = "bias"
elif "weight" in name:
# TODO: don't match quantizer.weight_proj
A_ = "weight"
else:
A_ = None
set_recursively(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
continue
if not is_used:
unused_weights.append(__UpperCamelCase )
logger.warning(f'''Unused weights: {unused_weights}''' )
def __snake_case ( __UpperCamelCase : Optional[int] ,__UpperCamelCase : Dict ,__UpperCamelCase : List[str] ,__UpperCamelCase : Union[str, Any] ,__UpperCamelCase : Optional[int] ):
"""simple docstring"""
A_ = full_name.split("conv_layers." )[-1]
A_ = name.split("." )
A_ = int(items[0] )
A_ = int(items[1] )
if type_id == 0:
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.bias.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape, (
f'''{full_name} has size {value.shape}, but {feature_extractor[layer_id].layer_norm.bias.data.shape} was'''
" found."
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor[layer_id].layer_norm.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(__UpperCamelCase )
@torch.no_grad()
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : str ,__UpperCamelCase : int=None ):
"""simple docstring"""
A_ = torch.load(__UpperCamelCase )
A_ = WavLMConfigOrig(checkpoint["cfg"] )
A_ = WavLMOrig(__UpperCamelCase )
model.load_state_dict(checkpoint["model"] )
model.eval()
if config_path is not None:
A_ = WavLMConfig.from_pretrained(__UpperCamelCase )
else:
A_ = WavLMConfig()
A_ = WavLMModel(__UpperCamelCase )
recursively_load_weights(__UpperCamelCase ,__UpperCamelCase )
hf_wavlm.save_pretrained(__UpperCamelCase )
if __name__ == "__main__":
__a :List[Any] = argparse.ArgumentParser()
parser.add_argument('--pytorch_dump_folder_path', default=None, type=str, help='Path to the output PyTorch model.')
parser.add_argument('--checkpoint_path', default=None, type=str, help='Path to fairseq checkpoint')
parser.add_argument('--config_path', default=None, type=str, help='Path to hf config.json of model to convert')
__a :Optional[int] = parser.parse_args()
convert_wavlm_checkpoint(args.checkpoint_path, args.pytorch_dump_folder_path, args.config_path) | 86 | 1 |
from typing import Dict
import numpy as np
import torch
from . import residue_constants as rc
from .tensor_utils import tensor_tree_map, tree_map
def __snake_case ( __UpperCamelCase : Dict[str, torch.Tensor] ):
"""simple docstring"""
A_ = []
A_ = []
A_ = []
for rt in rc.restypes:
A_ = rc.restype_name_to_atomaa_names[rc.restype_atoa[rt]]
restype_atomaa_to_atomaa_list.append([(rc.atom_order[name] if name else 0) for name in atom_names] )
A_ = {name: i for i, name in enumerate(__UpperCamelCase )}
restype_atomaa_to_atomaa_list.append(
[(atom_name_to_idxaa[name] if name in atom_name_to_idxaa else 0) for name in rc.atom_types] )
restype_atomaa_mask_list.append([(1.0 if name else 0.0) for name in atom_names] )
# Add dummy mapping for restype 'UNK'
restype_atomaa_to_atomaa_list.append([0] * 14 )
restype_atomaa_to_atomaa_list.append([0] * 37 )
restype_atomaa_mask_list.append([0.0] * 14 )
A_ = torch.tensor(
__UpperCamelCase ,dtype=torch.intaa ,device=protein["aatype"].device ,)
A_ = torch.tensor(
__UpperCamelCase ,dtype=torch.intaa ,device=protein["aatype"].device ,)
A_ = torch.tensor(
__UpperCamelCase ,dtype=torch.floataa ,device=protein["aatype"].device ,)
A_ = protein["aatype"].to(torch.long )
# create the mapping for (residx, atom14) --> atom37, i.e. an array
# with shape (num_res, 14) containing the atom37 indices for this protein
A_ = restype_atomaa_to_atomaa[protein_aatype]
A_ = restype_atomaa_mask[protein_aatype]
A_ = residx_atomaa_mask
A_ = residx_atomaa_to_atomaa.long()
# create the gather indices for mapping back
A_ = restype_atomaa_to_atomaa[protein_aatype]
A_ = residx_atomaa_to_atomaa.long()
# create the corresponding mask
A_ = torch.zeros([21, 37] ,dtype=torch.floataa ,device=protein["aatype"].device )
for restype, restype_letter in enumerate(rc.restypes ):
A_ = rc.restype_atoa[restype_letter]
A_ = rc.residue_atoms[restype_name]
for atom_name in atom_names:
A_ = rc.atom_order[atom_name]
A_ = 1
A_ = restype_atomaa_mask[protein_aatype]
A_ = residx_atomaa_mask
return protein
def __snake_case ( __UpperCamelCase : Dict[str, torch.Tensor] ):
"""simple docstring"""
A_ = tree_map(lambda __UpperCamelCase : torch.tensor(__UpperCamelCase ,device=batch["aatype"].device ) ,__UpperCamelCase ,np.ndarray )
A_ = tensor_tree_map(lambda __UpperCamelCase : np.array(__UpperCamelCase ) ,make_atomaa_masks(__UpperCamelCase ) )
return out | 86 |
def __snake_case ( __UpperCamelCase : list ,__UpperCamelCase : int = 0 ):
"""simple docstring"""
A_ = length or len(__UpperCamelCase )
A_ = False
for i in range(length - 1 ):
if list_data[i] > list_data[i + 1]:
A_ , A_ = list_data[i + 1], list_data[i]
A_ = True
return list_data if not swapped else bubble_sort(__UpperCamelCase ,length - 1 )
if __name__ == "__main__":
import doctest
doctest.testmod() | 86 | 1 |
def __snake_case ( __UpperCamelCase : int = 10**9 ):
"""simple docstring"""
A_ = 1
A_ = 2
A_ = 0
A_ = 0
A_ = 0
while perimeter <= max_perimeter:
perimeters_sum += perimeter
prev_value += 2 * value
value += prev_value
A_ = 2 * value + 2 if i % 2 == 0 else 2 * value - 2
i += 1
return perimeters_sum
if __name__ == "__main__":
print(F"{solution() = }") | 86 |
import pickle
import unittest
import torch
from accelerate import Accelerator
from accelerate.state import AcceleratorState
from accelerate.test_utils import require_cpu
@require_cpu
class _a ( unittest.TestCase ):
"""simple docstring"""
def __A ( self : List[str] ):
A_ = torch.nn.Linear(10 , 10 )
A_ = torch.optim.SGD(model.parameters() , 0.1 )
A_ = Accelerator()
A_ = accelerator.prepare(UpperCAmelCase )
try:
pickle.loads(pickle.dumps(UpperCAmelCase ) )
except Exception as e:
self.fail(f'''Accelerated optimizer pickling failed with {e}''' )
AcceleratorState._reset_state() | 86 | 1 |
from collections import OrderedDict
from typing import Mapping
from packaging import version
from ...configuration_utils import PretrainedConfig
from ...onnx import OnnxConfig
from ...utils import logging
from ..auto import CONFIG_MAPPING
__a :List[str] = logging.get_logger(__name__)
__a :Tuple = {
'microsoft/table-transformer-detection': (
'https://huggingface.co/microsoft/table-transformer-detection/resolve/main/config.json'
),
}
class _a ( snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Dict = 'table-transformer'
_lowerCamelCase : Any = ['past_key_values']
_lowerCamelCase : Optional[int] = {
'hidden_size': 'd_model',
'num_attention_heads': 'encoder_attention_heads',
}
def __init__( self : Dict , UpperCAmelCase : str=True , UpperCAmelCase : str=None , UpperCAmelCase : Optional[int]=3 , UpperCAmelCase : List[str]=100 , UpperCAmelCase : Union[str, Any]=6 , UpperCAmelCase : Optional[int]=2048 , UpperCAmelCase : Any=8 , UpperCAmelCase : Union[str, Any]=6 , UpperCAmelCase : str=2048 , UpperCAmelCase : Dict=8 , UpperCAmelCase : Tuple=0.0 , UpperCAmelCase : Tuple=0.0 , UpperCAmelCase : int=True , UpperCAmelCase : Any="relu" , UpperCAmelCase : int=256 , UpperCAmelCase : int=0.1 , UpperCAmelCase : Tuple=0.0 , UpperCAmelCase : str=0.0 , UpperCAmelCase : str=0.02 , UpperCAmelCase : List[str]=1.0 , UpperCAmelCase : Union[str, Any]=False , UpperCAmelCase : Dict="sine" , UpperCAmelCase : List[str]="resnet50" , UpperCAmelCase : str=True , UpperCAmelCase : Optional[Any]=False , UpperCAmelCase : Union[str, Any]=1 , UpperCAmelCase : Union[str, Any]=5 , UpperCAmelCase : List[Any]=2 , UpperCAmelCase : Dict=1 , UpperCAmelCase : Tuple=1 , UpperCAmelCase : Tuple=5 , UpperCAmelCase : Optional[Any]=2 , UpperCAmelCase : Optional[int]=0.1 , **UpperCAmelCase : int , ):
if backbone_config is not None and use_timm_backbone:
raise ValueError("You can't specify both `backbone_config` and `use_timm_backbone`." )
if not use_timm_backbone:
if backbone_config is None:
logger.info("`backbone_config` is `None`. Initializing the config with the default `ResNet` backbone." )
A_ = CONFIG_MAPPING["resnet"](out_features=["stage4"] )
elif isinstance(UpperCAmelCase , UpperCAmelCase ):
A_ = backbone_config.get("model_type" )
A_ = CONFIG_MAPPING[backbone_model_type]
A_ = config_class.from_dict(UpperCAmelCase )
# set timm attributes to None
A_ , A_ , A_ = None, None, None
A_ = use_timm_backbone
A_ = backbone_config
A_ = num_channels
A_ = num_queries
A_ = d_model
A_ = encoder_ffn_dim
A_ = encoder_layers
A_ = encoder_attention_heads
A_ = decoder_ffn_dim
A_ = decoder_layers
A_ = decoder_attention_heads
A_ = dropout
A_ = attention_dropout
A_ = activation_dropout
A_ = activation_function
A_ = init_std
A_ = init_xavier_std
A_ = encoder_layerdrop
A_ = decoder_layerdrop
A_ = encoder_layers
A_ = auxiliary_loss
A_ = position_embedding_type
A_ = backbone
A_ = use_pretrained_backbone
A_ = dilation
# Hungarian matcher
A_ = class_cost
A_ = bbox_cost
A_ = giou_cost
# Loss coefficients
A_ = mask_loss_coefficient
A_ = dice_loss_coefficient
A_ = bbox_loss_coefficient
A_ = giou_loss_coefficient
A_ = eos_coefficient
super().__init__(is_encoder_decoder=UpperCAmelCase , **UpperCAmelCase )
@property
def __A ( self : int ):
return self.encoder_attention_heads
@property
def __A ( self : int ):
return self.d_model
class _a ( snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Union[str, Any] = version.parse('1.11' )
@property
def __A ( self : Any ):
return OrderedDict(
[
("pixel_values", {0: "batch", 1: "num_channels", 2: "height", 3: "width"}),
("pixel_mask", {0: "batch"}),
] )
@property
def __A ( self : Any ):
return 1E-5
@property
def __A ( self : List[Any] ):
return 12 | 86 |
import argparse
import json
import os
import fairseq
import torch
from torch import nn
from transformers import (
SpeechaTextaConfig,
SpeechaTextaForCausalLM,
SpeechaTextaTokenizer,
SpeechEncoderDecoderConfig,
SpeechEncoderDecoderModel,
WavaVecaConfig,
WavaVecaFeatureExtractor,
WavaVecaModel,
logging,
)
logging.set_verbosity_info()
__a :List[str] = logging.get_logger(__name__)
__a :Optional[int] = {
'post_extract_proj': 'feature_projection.projection',
'encoder.pos_conv.0': 'encoder.pos_conv_embed.conv',
'self_attn.k_proj': 'encoder.layers.*.attention.k_proj',
'self_attn.v_proj': 'encoder.layers.*.attention.v_proj',
'self_attn.q_proj': 'encoder.layers.*.attention.q_proj',
'self_attn.out_proj': 'encoder.layers.*.attention.out_proj',
'self_attn_layer_norm': 'encoder.layers.*.layer_norm',
'fc1': 'encoder.layers.*.feed_forward.intermediate_dense',
'fc2': 'encoder.layers.*.feed_forward.output_dense',
'final_layer_norm': 'encoder.layers.*.final_layer_norm',
'encoder.layer_norm': 'encoder.layer_norm',
'w2v_model.layer_norm': 'feature_projection.layer_norm',
'quantizer.weight_proj': 'quantizer.weight_proj',
'quantizer.vars': 'quantizer.codevectors',
'project_q': 'project_q',
'final_proj': 'project_hid',
'w2v_encoder.proj': 'lm_head',
'mask_emb': 'masked_spec_embed',
}
__a :Any = [
'lm_head',
'quantizer.weight_proj',
'quantizer.codevectors',
'project_q',
'project_hid',
]
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : Optional[Any] ,__UpperCamelCase : Optional[int] ,__UpperCamelCase : List[str] ,__UpperCamelCase : Union[str, Any] ):
"""simple docstring"""
for attribute in key.split("." ):
A_ = getattr(__UpperCamelCase ,__UpperCamelCase )
if weight_type is not None:
A_ = getattr(__UpperCamelCase ,__UpperCamelCase ).shape
else:
A_ = hf_pointer.shape
assert hf_shape == value.shape, (
f'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
f''' {value.shape} for {full_name}'''
)
if weight_type == "weight":
A_ = value
elif weight_type == "weight_g":
A_ = value
elif weight_type == "weight_v":
A_ = value
elif weight_type == "bias":
A_ = value
else:
A_ = value
logger.info(f'''{key + "." + weight_type if weight_type is not None else ""} was initialized from {full_name}.''' )
def __snake_case ( __UpperCamelCase : Union[str, Any] ,__UpperCamelCase : Dict ):
"""simple docstring"""
A_ = []
A_ = fairseq_model.state_dict()
A_ = hf_model.feature_extractor
# if encoder has different dim to decoder -> use proj_weight
A_ = None
for name, value in fairseq_dict.items():
A_ = False
if "conv_layers" in name:
load_conv_layer(
__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,hf_model.config.feat_extract_norm == "group" ,)
A_ = True
elif name.split("." )[0] == "proj":
A_ = fairseq_model.proj
A_ = True
else:
for key, mapped_key in MAPPING.items():
if key in name or key.split("w2v_model." )[-1] == name.split("." )[0]:
A_ = True
if "*" in mapped_key:
A_ = name.split(__UpperCamelCase )[0].split("." )[-2]
A_ = mapped_key.replace("*" ,__UpperCamelCase )
if "weight_g" in name:
A_ = "weight_g"
elif "weight_v" in name:
A_ = "weight_v"
elif "bias" in name:
A_ = "bias"
elif "weight" in name:
A_ = "weight"
else:
A_ = None
set_recursively(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
continue
if not is_used:
unused_weights.append(__UpperCamelCase )
logger.warning(f'''Unused weights: {unused_weights}''' )
return proj_weight
def __snake_case ( __UpperCamelCase : str ,__UpperCamelCase : Optional[Any] ,__UpperCamelCase : int ,__UpperCamelCase : int ,__UpperCamelCase : Any ):
"""simple docstring"""
A_ = full_name.split("conv_layers." )[-1]
A_ = name.split("." )
A_ = int(items[0] )
A_ = int(items[1] )
if type_id == 0:
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.bias.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape, (
f'''{full_name} has size {value.shape}, but {feature_extractor[layer_id].layer_norm.bias.data.shape} was'''
" found."
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor[layer_id].layer_norm.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(__UpperCamelCase )
def __snake_case ( __UpperCamelCase : Optional[Any] ):
"""simple docstring"""
A_ , A_ = emb.weight.shape
A_ = nn.Linear(__UpperCamelCase ,__UpperCamelCase ,bias=__UpperCamelCase )
A_ = emb.weight.data
return lin_layer
def __snake_case ( __UpperCamelCase : Tuple ):
"""simple docstring"""
with open(__UpperCamelCase ,"r" ,encoding="utf-8" ) as f:
A_ = f.readlines()
A_ = [line.split(" " )[0] for line in lines]
A_ = len(__UpperCamelCase )
A_ = {
"<s>": 0,
"<pad>": 1,
"</s>": 2,
"<unk>": 3,
}
vocab_dict.update(dict(zip(__UpperCamelCase ,range(4 ,num_words + 4 ) ) ) )
return vocab_dict
@torch.no_grad()
def __snake_case ( __UpperCamelCase : List[str] ,__UpperCamelCase : Any ,__UpperCamelCase : List[Any] ,__UpperCamelCase : Union[str, Any] ,__UpperCamelCase : List[str] ,__UpperCamelCase : Optional[Any] ,__UpperCamelCase : Dict ,):
"""simple docstring"""
A_ = WavaVecaConfig.from_pretrained(__UpperCamelCase )
A_ = SpeechaTextaConfig.from_pretrained(
__UpperCamelCase ,vocab_size=__UpperCamelCase ,decoder_layers=__UpperCamelCase ,do_stable_layer_norm=__UpperCamelCase )
A_ = WavaVecaFeatureExtractor(
feature_size=1 ,sampling_rate=1_6000 ,padding_value=0 ,do_normalize=__UpperCamelCase ,return_attention_mask=__UpperCamelCase ,)
A_ , A_ , A_ = fairseq.checkpoint_utils.load_model_ensemble_and_task(
[checkpoint_path] ,arg_overrides={"data": "/".join(dict_path.split("/" )[:-1] )} )
A_ = model[0].eval()
# set weights for wav2vec2 encoder
A_ = WavaVecaModel(__UpperCamelCase )
A_ = recursively_load_weights_wavaveca(model.encoder ,__UpperCamelCase )
A_ = SpeechaTextaForCausalLM(__UpperCamelCase )
A_ , A_ = hf_decoder.model.decoder.load_state_dict(model.decoder.state_dict() ,strict=__UpperCamelCase )
# set output linear layer
unexpected_keys.remove("embed_out" )
A_ = nn.Parameter(model.decoder.embed_out.detach() )
# layer norm is init to identity matrix so leaving it is fine
logger.warning(f'''The following keys are missing when loading the decoder weights: {missing_keys}''' )
logger.warning(f'''The following keys are unexpected when loading the decoder weights: {unexpected_keys}''' )
A_ = SpeechEncoderDecoderModel(encoder=__UpperCamelCase ,decoder=__UpperCamelCase )
A_ = False
# add projection layer
A_ = nn.Parameter(projection_layer.weight )
A_ = nn.Parameter(projection_layer.bias )
A_ = create_vocab_dict(__UpperCamelCase )
with open(os.path.join(__UpperCamelCase ,"vocab.json" ) ,"w" ) as fp:
json.dump(__UpperCamelCase ,__UpperCamelCase )
A_ = SpeechaTextaTokenizer(os.path.join(__UpperCamelCase ,"vocab.json" ) )
tokenizer.save_pretrained(__UpperCamelCase )
A_ = hf_wavavec.config.to_dict()
A_ = tokenizer.pad_token_id
A_ = tokenizer.bos_token_id
A_ = tokenizer.eos_token_id
A_ = "speech_to_text_2"
A_ = "wav2vec2"
A_ = SpeechEncoderDecoderConfig.from_dict(__UpperCamelCase )
hf_wavavec.save_pretrained(__UpperCamelCase )
feature_extractor.save_pretrained(__UpperCamelCase )
if __name__ == "__main__":
__a :int = argparse.ArgumentParser()
parser.add_argument('--pytorch_dump_folder_path', default=None, type=str, help='Path to the output PyTorch model.')
parser.add_argument('--checkpoint_path', default=None, type=str, help='Path to fairseq checkpoint')
parser.add_argument('--dict_path', default=None, type=str, help='Path to dict of fine-tuned model')
parser.add_argument(
'--encoder_config_path',
default='facebook/wav2vec2-large-lv60',
type=str,
help='Path to hf encoder wav2vec2 checkpoint config',
)
parser.add_argument(
'--decoder_config_path',
default='facebook/s2t-small-mustc-en-fr-st',
type=str,
help='Path to hf decoder s2t checkpoint config',
)
parser.add_argument('--vocab_size', default=1_0224, type=int, help='Vocab size of decoder')
parser.add_argument('--num_decoder_layers', default=7, type=int, help='Number of decoder layers')
__a :Tuple = parser.parse_args()
convert_wavaveca_checkpoint(
args.checkpoint_path,
args.pytorch_dump_folder_path,
args.dict_path,
encoder_config_path=args.encoder_config_path,
decoder_config_path=args.decoder_config_path,
vocab_size=args.vocab_size,
num_decoder_layers=args.num_decoder_layers,
) | 86 | 1 |
import math
from ...configuration_utils import PretrainedConfig
from ...utils import logging
__a :Dict = logging.get_logger(__name__)
__a :Optional[Any] = {
'facebook/data2vec-base-960h': 'https://huggingface.co/facebook/data2vec-audio-base-960h/resolve/main/config.json',
# See all Data2VecAudio models at https://huggingface.co/models?filter=data2vec-audio
}
class _a ( snake_case_ ):
"""simple docstring"""
_lowerCamelCase : int = 'data2vec-audio'
def __init__( self : Dict , UpperCAmelCase : List[Any]=32 , UpperCAmelCase : Optional[int]=768 , UpperCAmelCase : int=12 , UpperCAmelCase : int=12 , UpperCAmelCase : Optional[Any]=3072 , UpperCAmelCase : Dict="gelu" , UpperCAmelCase : List[str]=0.1 , UpperCAmelCase : Any=0.1 , UpperCAmelCase : Union[str, Any]=0.1 , UpperCAmelCase : List[str]=0.0 , UpperCAmelCase : Optional[Any]=0.1 , UpperCAmelCase : Dict=0.1 , UpperCAmelCase : Any=0.02 , UpperCAmelCase : int=1E-5 , UpperCAmelCase : Optional[Any]="gelu" , UpperCAmelCase : Union[str, Any]=(512, 512, 512, 512, 512, 512, 512) , UpperCAmelCase : Optional[Any]=(5, 2, 2, 2, 2, 2, 2) , UpperCAmelCase : Optional[int]=(10, 3, 3, 3, 3, 2, 2) , UpperCAmelCase : int=False , UpperCAmelCase : Optional[int]=16 , UpperCAmelCase : List[Any]=19 , UpperCAmelCase : Tuple=5 , UpperCAmelCase : List[str]=0.05 , UpperCAmelCase : int=10 , UpperCAmelCase : List[Any]=2 , UpperCAmelCase : Optional[Any]=0.0 , UpperCAmelCase : int=10 , UpperCAmelCase : List[str]=0 , UpperCAmelCase : List[Any]="sum" , UpperCAmelCase : int=False , UpperCAmelCase : List[str]=False , UpperCAmelCase : Tuple=256 , UpperCAmelCase : Union[str, Any]=(512, 512, 512, 512, 1500) , UpperCAmelCase : Dict=(5, 3, 3, 1, 1) , UpperCAmelCase : Union[str, Any]=(1, 2, 3, 1, 1) , UpperCAmelCase : Optional[int]=512 , UpperCAmelCase : int=0 , UpperCAmelCase : Optional[int]=1 , UpperCAmelCase : Dict=2 , UpperCAmelCase : int=False , UpperCAmelCase : Optional[Any]=3 , UpperCAmelCase : Optional[Any]=2 , UpperCAmelCase : List[str]=3 , UpperCAmelCase : str=None , **UpperCAmelCase : Union[str, Any] , ):
super().__init__(**UpperCAmelCase , pad_token_id=UpperCAmelCase , bos_token_id=UpperCAmelCase , eos_token_id=UpperCAmelCase )
A_ = hidden_size
A_ = feat_extract_activation
A_ = list(UpperCAmelCase )
A_ = list(UpperCAmelCase )
A_ = list(UpperCAmelCase )
A_ = conv_bias
A_ = num_conv_pos_embeddings
A_ = num_conv_pos_embedding_groups
A_ = conv_pos_kernel_size
A_ = len(self.conv_dim )
A_ = num_hidden_layers
A_ = intermediate_size
A_ = hidden_act
A_ = num_attention_heads
A_ = hidden_dropout
A_ = attention_dropout
A_ = activation_dropout
A_ = feat_proj_dropout
A_ = final_dropout
A_ = layerdrop
A_ = layer_norm_eps
A_ = initializer_range
A_ = vocab_size
A_ = use_weighted_layer_sum
if (
(len(self.conv_stride ) != self.num_feat_extract_layers)
or (len(self.conv_kernel ) != self.num_feat_extract_layers)
or (len(self.conv_dim ) != self.num_feat_extract_layers)
):
raise ValueError(
"Configuration for convolutional layers is incorrect. It is required that `len(config.conv_dim)` =="
" `len(config.conv_stride)` == `len(config.conv_kernel)`, but is `len(config.conv_dim) ="
f''' {len(self.conv_dim )}`, `len(config.conv_stride) = {len(self.conv_stride )}`,'''
f''' `len(config.conv_kernel) = {len(self.conv_kernel )}`.''' )
# fine-tuning config parameters for SpecAugment: https://arxiv.org/abs/1904.08779
A_ = mask_time_prob
A_ = mask_time_length
A_ = mask_time_min_masks
A_ = mask_feature_prob
A_ = mask_feature_length
A_ = mask_feature_min_masks
# ctc loss
A_ = ctc_loss_reduction
A_ = ctc_zero_infinity
# adapter
A_ = add_adapter
A_ = adapter_kernel_size
A_ = adapter_stride
A_ = num_adapter_layers
A_ = output_hidden_size or hidden_size
# SequenceClassification-specific parameter. Feel free to ignore for other classes.
A_ = classifier_proj_size
# XVector-specific parameters. Feel free to ignore for other classes.
A_ = list(UpperCAmelCase )
A_ = list(UpperCAmelCase )
A_ = list(UpperCAmelCase )
A_ = xvector_output_dim
@property
def __A ( self : Optional[Any] ):
return math.prod(self.conv_stride ) | 86 |
from typing import Any, Dict, List, Union
from ..utils import add_end_docstrings, is_torch_available, is_vision_available, logging, requires_backends
from .base import PIPELINE_INIT_ARGS, Pipeline
if is_vision_available():
from ..image_utils import load_image
if is_torch_available():
import torch
from ..models.auto.modeling_auto import MODEL_FOR_OBJECT_DETECTION_MAPPING, MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING
__a :str = logging.get_logger(__name__)
__a :Any = Dict[str, Any]
__a :int = List[Prediction]
@add_end_docstrings(snake_case_ )
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Tuple , *UpperCAmelCase : Optional[int] , **UpperCAmelCase : Optional[Any] ):
super().__init__(*UpperCAmelCase , **UpperCAmelCase )
if self.framework == "tf":
raise ValueError(f'''The {self.__class__} is only available in PyTorch.''' )
requires_backends(self , "vision" )
self.check_model_type(
dict(MODEL_FOR_OBJECT_DETECTION_MAPPING.items() + MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING.items() ) )
def __A ( self : str , **UpperCAmelCase : str ):
A_ = {}
if "threshold" in kwargs:
A_ = kwargs["threshold"]
return {}, {}, postprocess_kwargs
def __call__( self : Union[str, Any] , *UpperCAmelCase : Union[str, Any] , **UpperCAmelCase : Optional[Any] ):
return super().__call__(*UpperCAmelCase , **UpperCAmelCase )
def __A ( self : str , UpperCAmelCase : Any ):
A_ = load_image(UpperCAmelCase )
A_ = torch.IntTensor([[image.height, image.width]] )
A_ = self.image_processor(images=[image] , return_tensors="pt" )
if self.tokenizer is not None:
A_ = self.tokenizer(text=inputs["words"] , boxes=inputs["boxes"] , return_tensors="pt" )
A_ = target_size
return inputs
def __A ( self : Optional[Any] , UpperCAmelCase : Optional[int] ):
A_ = model_inputs.pop("target_size" )
A_ = self.model(**UpperCAmelCase )
A_ = outputs.__class__({"target_size": target_size, **outputs} )
if self.tokenizer is not None:
A_ = model_inputs["bbox"]
return model_outputs
def __A ( self : str , UpperCAmelCase : Optional[int] , UpperCAmelCase : Union[str, Any]=0.9 ):
A_ = model_outputs["target_size"]
if self.tokenizer is not None:
# This is a LayoutLMForTokenClassification variant.
# The OCR got the boxes and the model classified the words.
A_ , A_ = target_size[0].tolist()
def unnormalize(UpperCAmelCase : Any ):
return self._get_bounding_box(
torch.Tensor(
[
(width * bbox[0] / 1000),
(height * bbox[1] / 1000),
(width * bbox[2] / 1000),
(height * bbox[3] / 1000),
] ) )
A_ , A_ = model_outputs["logits"].squeeze(0 ).softmax(dim=-1 ).max(dim=-1 )
A_ = [self.model.config.idalabel[prediction] for prediction in classes.tolist()]
A_ = [unnormalize(UpperCAmelCase ) for bbox in model_outputs["bbox"].squeeze(0 )]
A_ = ["score", "label", "box"]
A_ = [dict(zip(UpperCAmelCase , UpperCAmelCase ) ) for vals in zip(scores.tolist() , UpperCAmelCase , UpperCAmelCase ) if vals[0] > threshold]
else:
# This is a regular ForObjectDetectionModel
A_ = self.image_processor.post_process_object_detection(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase )
A_ = raw_annotations[0]
A_ = raw_annotation["scores"]
A_ = raw_annotation["labels"]
A_ = raw_annotation["boxes"]
A_ = scores.tolist()
A_ = [self.model.config.idalabel[label.item()] for label in labels]
A_ = [self._get_bounding_box(UpperCAmelCase ) for box in boxes]
# {"scores": [...], ...} --> [{"score":x, ...}, ...]
A_ = ["score", "label", "box"]
A_ = [
dict(zip(UpperCAmelCase , UpperCAmelCase ) )
for vals in zip(raw_annotation["scores"] , raw_annotation["labels"] , raw_annotation["boxes"] )
]
return annotation
def __A ( self : Tuple , UpperCAmelCase : "torch.Tensor" ):
if self.framework != "pt":
raise ValueError("The ObjectDetectionPipeline is only available in PyTorch." )
A_ , A_ , A_ , A_ = box.int().tolist()
A_ = {
"xmin": xmin,
"ymin": ymin,
"xmax": xmax,
"ymax": ymax,
}
return bbox | 86 | 1 |
from __future__ import annotations
class _a :
"""simple docstring"""
def __init__( self : Dict , UpperCAmelCase : int ):
A_ = order
# a_{0} ... a_{k}
A_ = [1.0] + [0.0] * order
# b_{0} ... b_{k}
A_ = [1.0] + [0.0] * order
# x[n-1] ... x[n-k]
A_ = [0.0] * self.order
# y[n-1] ... y[n-k]
A_ = [0.0] * self.order
def __A ( self : Union[str, Any] , UpperCAmelCase : list[float] , UpperCAmelCase : list[float] ):
if len(UpperCAmelCase ) < self.order:
A_ = [1.0, *a_coeffs]
if len(UpperCAmelCase ) != self.order + 1:
A_ = (
f'''Expected a_coeffs to have {self.order + 1} elements '''
f'''for {self.order}-order filter, got {len(UpperCAmelCase )}'''
)
raise ValueError(UpperCAmelCase )
if len(UpperCAmelCase ) != self.order + 1:
A_ = (
f'''Expected b_coeffs to have {self.order + 1} elements '''
f'''for {self.order}-order filter, got {len(UpperCAmelCase )}'''
)
raise ValueError(UpperCAmelCase )
A_ = a_coeffs
A_ = b_coeffs
def __A ( self : Tuple , UpperCAmelCase : float ):
A_ = 0.0
# Start at index 1 and do index 0 at the end.
for i in range(1 , self.order + 1 ):
result += (
self.b_coeffs[i] * self.input_history[i - 1]
- self.a_coeffs[i] * self.output_history[i - 1]
)
A_ = (result + self.b_coeffs[0] * sample) / self.a_coeffs[0]
A_ = self.input_history[:-1]
A_ = self.output_history[:-1]
A_ = sample
A_ = result
return result | 86 |
import inspect
from typing import List, Optional, Tuple, Union
import numpy as np
import PIL
import torch
import torch.utils.checkpoint
from ...models import UNetaDModel, VQModel
from ...schedulers import (
DDIMScheduler,
DPMSolverMultistepScheduler,
EulerAncestralDiscreteScheduler,
EulerDiscreteScheduler,
LMSDiscreteScheduler,
PNDMScheduler,
)
from ...utils import PIL_INTERPOLATION, randn_tensor
from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
def __snake_case ( __UpperCamelCase : Dict ):
"""simple docstring"""
A_ , A_ = image.size
A_ , A_ = (x - x % 32 for x in (w, h)) # resize to integer multiple of 32
A_ = image.resize((w, h) ,resample=PIL_INTERPOLATION["lanczos"] )
A_ = np.array(__UpperCamelCase ).astype(np.floataa ) / 255.0
A_ = image[None].transpose(0 ,3 ,1 ,2 )
A_ = torch.from_numpy(__UpperCamelCase )
return 2.0 * image - 1.0
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Union[str, Any] , UpperCAmelCase : VQModel , UpperCAmelCase : UNetaDModel , UpperCAmelCase : Union[
DDIMScheduler,
PNDMScheduler,
LMSDiscreteScheduler,
EulerDiscreteScheduler,
EulerAncestralDiscreteScheduler,
DPMSolverMultistepScheduler,
] , ):
super().__init__()
self.register_modules(vqvae=UpperCAmelCase , unet=UpperCAmelCase , scheduler=UpperCAmelCase )
@torch.no_grad()
def __call__( self : int , UpperCAmelCase : Union[torch.Tensor, PIL.Image.Image] = None , UpperCAmelCase : Optional[int] = 1 , UpperCAmelCase : Optional[int] = 100 , UpperCAmelCase : Optional[float] = 0.0 , UpperCAmelCase : Optional[Union[torch.Generator, List[torch.Generator]]] = None , UpperCAmelCase : Optional[str] = "pil" , UpperCAmelCase : bool = True , ):
if isinstance(UpperCAmelCase , PIL.Image.Image ):
A_ = 1
elif isinstance(UpperCAmelCase , torch.Tensor ):
A_ = image.shape[0]
else:
raise ValueError(f'''`image` has to be of type `PIL.Image.Image` or `torch.Tensor` but is {type(UpperCAmelCase )}''' )
if isinstance(UpperCAmelCase , PIL.Image.Image ):
A_ = preprocess(UpperCAmelCase )
A_ , A_ = image.shape[-2:]
# in_channels should be 6: 3 for latents, 3 for low resolution image
A_ = (batch_size, self.unet.config.in_channels // 2, height, width)
A_ = next(self.unet.parameters() ).dtype
A_ = randn_tensor(UpperCAmelCase , generator=UpperCAmelCase , device=self.device , dtype=UpperCAmelCase )
A_ = image.to(device=self.device , dtype=UpperCAmelCase )
# set timesteps and move to the correct device
self.scheduler.set_timesteps(UpperCAmelCase , device=self.device )
A_ = self.scheduler.timesteps
# scale the initial noise by the standard deviation required by the scheduler
A_ = latents * self.scheduler.init_noise_sigma
# prepare extra kwargs for the scheduler step, since not all schedulers have the same signature.
# eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
# eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
# and should be between [0, 1]
A_ = "eta" in set(inspect.signature(self.scheduler.step ).parameters.keys() )
A_ = {}
if accepts_eta:
A_ = eta
for t in self.progress_bar(UpperCAmelCase ):
# concat latents and low resolution image in the channel dimension.
A_ = torch.cat([latents, image] , dim=1 )
A_ = self.scheduler.scale_model_input(UpperCAmelCase , UpperCAmelCase )
# predict the noise residual
A_ = self.unet(UpperCAmelCase , UpperCAmelCase ).sample
# compute the previous noisy sample x_t -> x_t-1
A_ = self.scheduler.step(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase , **UpperCAmelCase ).prev_sample
# decode the image latents with the VQVAE
A_ = self.vqvae.decode(UpperCAmelCase ).sample
A_ = torch.clamp(UpperCAmelCase , -1.0 , 1.0 )
A_ = image / 2 + 0.5
A_ = image.cpu().permute(0 , 2 , 3 , 1 ).numpy()
if output_type == "pil":
A_ = self.numpy_to_pil(UpperCAmelCase )
if not return_dict:
return (image,)
return ImagePipelineOutput(images=UpperCAmelCase ) | 86 | 1 |
import importlib.metadata
import operator
import re
import sys
from typing import Optional
from packaging import version
__a :Optional[int] = {
'<': operator.lt,
'<=': operator.le,
'==': operator.eq,
'!=': operator.ne,
'>=': operator.ge,
'>': operator.gt,
}
def __snake_case ( __UpperCamelCase : Tuple ,__UpperCamelCase : Dict ,__UpperCamelCase : Optional[Any] ,__UpperCamelCase : Union[str, Any] ,__UpperCamelCase : Any ,__UpperCamelCase : Any ):
"""simple docstring"""
if got_ver is None or want_ver is None:
raise ValueError(
f'''Unable to compare versions for {requirement}: need={want_ver} found={got_ver}. This is unusual. Consider'''
f''' reinstalling {pkg}.''' )
if not ops[op](version.parse(__UpperCamelCase ) ,version.parse(__UpperCamelCase ) ):
raise ImportError(
f'''{requirement} is required for a normal functioning of this module, but found {pkg}=={got_ver}.{hint}''' )
def __snake_case ( __UpperCamelCase : str ,__UpperCamelCase : Optional[str] = None ):
"""simple docstring"""
A_ = f'''\n{hint}''' if hint is not None else ""
# non-versioned check
if re.match(R"^[\w_\-\d]+$" ,__UpperCamelCase ):
A_ , A_ , A_ = requirement, None, None
else:
A_ = re.findall(R"^([^!=<>\s]+)([\s!=<>]{1,2}.+)" ,__UpperCamelCase )
if not match:
raise ValueError(
"requirement needs to be in the pip package format, .e.g., package_a==1.23, or package_b>=1.23, but"
f''' got {requirement}''' )
A_ , A_ = match[0]
A_ = want_full.split("," ) # there could be multiple requirements
A_ = {}
for w in want_range:
A_ = re.findall(R"^([\s!=<>]{1,2})(.+)" ,__UpperCamelCase )
if not match:
raise ValueError(
"requirement needs to be in the pip package format, .e.g., package_a==1.23, or package_b>=1.23,"
f''' but got {requirement}''' )
A_ , A_ = match[0]
A_ = want_ver
if op not in ops:
raise ValueError(f'''{requirement}: need one of {list(ops.keys() )}, but got {op}''' )
# special case
if pkg == "python":
A_ = ".".join([str(__UpperCamelCase ) for x in sys.version_info[:3]] )
for op, want_ver in wanted.items():
_compare_versions(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
return
# check if any version is installed
try:
A_ = importlib.metadata.version(__UpperCamelCase )
except importlib.metadata.PackageNotFoundError:
raise importlib.metadata.PackageNotFoundError(
f'''The \'{requirement}\' distribution was not found and is required by this application. {hint}''' )
# check that the right version is installed if version number or a range was provided
if want_ver is not None:
for op, want_ver in wanted.items():
_compare_versions(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
def __snake_case ( __UpperCamelCase : List[Any] ):
"""simple docstring"""
A_ = "Try: pip install transformers -U or pip install -e '.[dev]' if you're working with git main"
return require_version(__UpperCamelCase ,__UpperCamelCase ) | 86 |
__a :Optional[int] = [sum(int(c, 10) ** 2 for c in i.__str__()) for i in range(10_0000)]
def __snake_case ( __UpperCamelCase : int ):
"""simple docstring"""
A_ = 0
while number:
# Increased Speed Slightly by checking every 5 digits together.
sum_of_digits_squared += DIGITS_SQUARED[number % 10_0000]
number //= 10_0000
return sum_of_digits_squared
# There are 2 Chains made,
# One ends with 89 with the chain member 58 being the one which when declared first,
# there will be the least number of iterations for all the members to be checked.
# The other one ends with 1 and has only one element 1.
# So 58 and 1 are chosen to be declared at the starting.
# Changed dictionary to an array to quicken the solution
__a :list[bool | None] = [None] * 1000_0000
__a :Optional[Any] = True
__a :List[Any] = False
def __snake_case ( __UpperCamelCase : int ):
"""simple docstring"""
if CHAINS[number - 1] is not None:
return CHAINS[number - 1] # type: ignore
A_ = chain(next_number(__UpperCamelCase ) )
A_ = number_chain
while number < 1000_0000:
A_ = number_chain
number *= 10
return number_chain
def __snake_case ( __UpperCamelCase : int = 1000_0000 ):
"""simple docstring"""
for i in range(1 ,__UpperCamelCase ):
if CHAINS[i] is None:
chain(i + 1 )
return CHAINS[:number].count(__UpperCamelCase )
if __name__ == "__main__":
import doctest
doctest.testmod()
print(F"{solution() = }") | 86 | 1 |
# Author: OMKAR PATHAK, Nwachukwu Chidiebere
# Use a Python dictionary to construct the graph.
from __future__ import annotations
from pprint import pformat
from typing import Generic, TypeVar
__a :str = TypeVar('T')
class _a ( Generic[T] ):
"""simple docstring"""
def __init__( self : Tuple , UpperCAmelCase : bool = True ):
A_ = {} # dictionary of lists
A_ = directed
def __A ( self : List[Any] , UpperCAmelCase : T , UpperCAmelCase : T ):
if not self.directed: # For undirected graphs
# if both source vertex and destination vertex are both present in the
# adjacency list, add destination vertex to source vertex list of adjacent
# vertices and add source vertex to destination vertex list of adjacent
# vertices.
if source_vertex in self.adj_list and destination_vertex in self.adj_list:
self.adj_list[source_vertex].append(UpperCAmelCase )
self.adj_list[destination_vertex].append(UpperCAmelCase )
# if only source vertex is present in adjacency list, add destination vertex
# to source vertex list of adjacent vertices, then create a new vertex with
# destination vertex as key and assign a list containing the source vertex
# as it's first adjacent vertex.
elif source_vertex in self.adj_list:
self.adj_list[source_vertex].append(UpperCAmelCase )
A_ = [source_vertex]
# if only destination vertex is present in adjacency list, add source vertex
# to destination vertex list of adjacent vertices, then create a new vertex
# with source vertex as key and assign a list containing the source vertex
# as it's first adjacent vertex.
elif destination_vertex in self.adj_list:
self.adj_list[destination_vertex].append(UpperCAmelCase )
A_ = [destination_vertex]
# if both source vertex and destination vertex are not present in adjacency
# list, create a new vertex with source vertex as key and assign a list
# containing the destination vertex as it's first adjacent vertex also
# create a new vertex with destination vertex as key and assign a list
# containing the source vertex as it's first adjacent vertex.
else:
A_ = [destination_vertex]
A_ = [source_vertex]
else: # For directed graphs
# if both source vertex and destination vertex are present in adjacency
# list, add destination vertex to source vertex list of adjacent vertices.
if source_vertex in self.adj_list and destination_vertex in self.adj_list:
self.adj_list[source_vertex].append(UpperCAmelCase )
# if only source vertex is present in adjacency list, add destination
# vertex to source vertex list of adjacent vertices and create a new vertex
# with destination vertex as key, which has no adjacent vertex
elif source_vertex in self.adj_list:
self.adj_list[source_vertex].append(UpperCAmelCase )
A_ = []
# if only destination vertex is present in adjacency list, create a new
# vertex with source vertex as key and assign a list containing destination
# vertex as first adjacent vertex
elif destination_vertex in self.adj_list:
A_ = [destination_vertex]
# if both source vertex and destination vertex are not present in adjacency
# list, create a new vertex with source vertex as key and a list containing
# destination vertex as it's first adjacent vertex. Then create a new vertex
# with destination vertex as key, which has no adjacent vertex
else:
A_ = [destination_vertex]
A_ = []
return self
def __repr__( self : Optional[Any] ):
return pformat(self.adj_list ) | 86 |
from typing import TYPE_CHECKING
from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_tf_available, is_torch_available
__a :List[Any] = {
'configuration_tapas': ['TAPAS_PRETRAINED_CONFIG_ARCHIVE_MAP', 'TapasConfig'],
'tokenization_tapas': ['TapasTokenizer'],
}
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Any = [
'TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST',
'TapasForMaskedLM',
'TapasForQuestionAnswering',
'TapasForSequenceClassification',
'TapasModel',
'TapasPreTrainedModel',
'load_tf_weights_in_tapas',
]
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Dict = [
'TF_TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST',
'TFTapasForMaskedLM',
'TFTapasForQuestionAnswering',
'TFTapasForSequenceClassification',
'TFTapasModel',
'TFTapasPreTrainedModel',
]
if TYPE_CHECKING:
from .configuration_tapas import TAPAS_PRETRAINED_CONFIG_ARCHIVE_MAP, TapasConfig
from .tokenization_tapas import TapasTokenizer
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tapas import (
TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST,
TapasForMaskedLM,
TapasForQuestionAnswering,
TapasForSequenceClassification,
TapasModel,
TapasPreTrainedModel,
load_tf_weights_in_tapas,
)
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tf_tapas import (
TF_TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST,
TFTapasForMaskedLM,
TFTapasForQuestionAnswering,
TFTapasForSequenceClassification,
TFTapasModel,
TFTapasPreTrainedModel,
)
else:
import sys
__a :str = _LazyModule(__name__, globals()['__file__'], _import_structure, module_spec=__spec__) | 86 | 1 |
from ...configuration_utils import PretrainedConfig
__a :str = {
'google/tapas-base-finetuned-sqa': (
'https://huggingface.co/google/tapas-base-finetuned-sqa/resolve/main/config.json'
),
'google/tapas-base-finetuned-wtq': (
'https://huggingface.co/google/tapas-base-finetuned-wtq/resolve/main/config.json'
),
'google/tapas-base-finetuned-wikisql-supervised': (
'https://huggingface.co/google/tapas-base-finetuned-wikisql-supervised/resolve/main/config.json'
),
'google/tapas-base-finetuned-tabfact': (
'https://huggingface.co/google/tapas-base-finetuned-tabfact/resolve/main/config.json'
),
}
class _a ( snake_case_ ):
"""simple docstring"""
_lowerCamelCase : int = 'tapas'
def __init__( self : Optional[Any] , UpperCAmelCase : Dict=30522 , UpperCAmelCase : List[str]=768 , UpperCAmelCase : Tuple=12 , UpperCAmelCase : List[Any]=12 , UpperCAmelCase : Dict=3072 , UpperCAmelCase : Tuple="gelu" , UpperCAmelCase : List[Any]=0.1 , UpperCAmelCase : int=0.1 , UpperCAmelCase : Tuple=1024 , UpperCAmelCase : Dict=[3, 256, 256, 2, 256, 256, 10] , UpperCAmelCase : str=0.02 , UpperCAmelCase : Union[str, Any]=1E-12 , UpperCAmelCase : List[str]=0 , UpperCAmelCase : Optional[int]=10.0 , UpperCAmelCase : int=0 , UpperCAmelCase : Any=1.0 , UpperCAmelCase : List[Any]=None , UpperCAmelCase : Union[str, Any]=1.0 , UpperCAmelCase : List[Any]=False , UpperCAmelCase : int=None , UpperCAmelCase : List[str]=1.0 , UpperCAmelCase : Optional[int]=1.0 , UpperCAmelCase : List[Any]=False , UpperCAmelCase : Tuple=False , UpperCAmelCase : List[str]="ratio" , UpperCAmelCase : List[str]=None , UpperCAmelCase : List[str]=None , UpperCAmelCase : Union[str, Any]=64 , UpperCAmelCase : Optional[Any]=32 , UpperCAmelCase : Tuple=False , UpperCAmelCase : List[str]=True , UpperCAmelCase : Optional[int]=False , UpperCAmelCase : Union[str, Any]=False , UpperCAmelCase : Tuple=True , UpperCAmelCase : Union[str, Any]=False , UpperCAmelCase : List[Any]=None , UpperCAmelCase : int=None , **UpperCAmelCase : str , ):
super().__init__(pad_token_id=UpperCAmelCase , **UpperCAmelCase )
# BERT hyperparameters (with updated max_position_embeddings and type_vocab_sizes)
A_ = vocab_size
A_ = hidden_size
A_ = num_hidden_layers
A_ = num_attention_heads
A_ = hidden_act
A_ = intermediate_size
A_ = hidden_dropout_prob
A_ = attention_probs_dropout_prob
A_ = max_position_embeddings
A_ = type_vocab_sizes
A_ = initializer_range
A_ = layer_norm_eps
# Fine-tuning task hyperparameters
A_ = positive_label_weight
A_ = num_aggregation_labels
A_ = aggregation_loss_weight
A_ = use_answer_as_supervision
A_ = answer_loss_importance
A_ = use_normalized_answer_loss
A_ = huber_loss_delta
A_ = temperature
A_ = aggregation_temperature
A_ = use_gumbel_for_cells
A_ = use_gumbel_for_aggregation
A_ = average_approximation_function
A_ = cell_selection_preference
A_ = answer_loss_cutoff
A_ = max_num_rows
A_ = max_num_columns
A_ = average_logits_per_cell
A_ = select_one_column
A_ = allow_empty_column_selection
A_ = init_cell_selection_weights_to_zero
A_ = reset_position_index_per_cell
A_ = disable_per_token_loss
# Aggregation hyperparameters
A_ = aggregation_labels
A_ = no_aggregation_label_index
if isinstance(self.aggregation_labels , UpperCAmelCase ):
A_ = {int(UpperCAmelCase ): v for k, v in aggregation_labels.items()} | 86 |
# Lint as: python3
import sys
from collections.abc import Mapping
from typing import TYPE_CHECKING, Dict, Optional
import numpy as np
import pyarrow as pa
from .. import config
from ..utils.logging import get_logger
from ..utils.py_utils import map_nested
from .formatting import TensorFormatter
if TYPE_CHECKING:
import jax
import jaxlib
__a :List[Any] = get_logger()
__a :Optional[dict] = None
class _a ( TensorFormatter[Mapping, 'jax.Array', Mapping] ):
"""simple docstring"""
def __init__( self : str , UpperCAmelCase : int=None , UpperCAmelCase : List[str]=None , **UpperCAmelCase : List[Any] ):
super().__init__(features=UpperCAmelCase )
import jax
from jaxlib.xla_client import Device
if isinstance(UpperCAmelCase , UpperCAmelCase ):
raise ValueError(
f'''Expected {device} to be a `str` not {type(UpperCAmelCase )}, as `jaxlib.xla_extension.Device` '''
"is not serializable neither with `pickle` nor with `dill`. Instead you can surround "
"the device with `str()` to get its string identifier that will be internally mapped "
"to the actual `jaxlib.xla_extension.Device`." )
A_ = device if isinstance(UpperCAmelCase , UpperCAmelCase ) else str(jax.devices()[0] )
# using global variable since `jaxlib.xla_extension.Device` is not serializable neither
# with `pickle` nor with `dill`, so we need to use a global variable instead
global DEVICE_MAPPING
if DEVICE_MAPPING is None:
A_ = self._map_devices_to_str()
if self.device not in list(DEVICE_MAPPING.keys() ):
logger.warning(
f'''Device with string identifier {self.device} not listed among the available '''
f'''devices: {list(DEVICE_MAPPING.keys() )}, so falling back to the default '''
f'''device: {str(jax.devices()[0] )}.''' )
A_ = str(jax.devices()[0] )
A_ = jnp_array_kwargs
@staticmethod
def __A ( ):
import jax
return {str(UpperCAmelCase ): device for device in jax.devices()}
def __A ( self : Optional[int] , UpperCAmelCase : int ):
import jax
import jax.numpy as jnp
if isinstance(UpperCAmelCase , UpperCAmelCase ) and column:
if all(
isinstance(UpperCAmelCase , jax.Array ) and x.shape == column[0].shape and x.dtype == column[0].dtype for x in column ):
return jnp.stack(UpperCAmelCase , axis=0 )
return column
def __A ( self : List[str] , UpperCAmelCase : str ):
import jax
import jax.numpy as jnp
if isinstance(UpperCAmelCase , (str, bytes, type(UpperCAmelCase )) ):
return value
elif isinstance(UpperCAmelCase , (np.character, np.ndarray) ) and np.issubdtype(value.dtype , np.character ):
return value.tolist()
A_ = {}
if isinstance(UpperCAmelCase , (np.number, np.ndarray) ) and np.issubdtype(value.dtype , np.integer ):
# the default int precision depends on the jax config
# see https://jax.readthedocs.io/en/latest/notebooks/Common_Gotchas_in_JAX.html#double-64bit-precision
if jax.config.jax_enable_xaa:
A_ = {"dtype": jnp.intaa}
else:
A_ = {"dtype": jnp.intaa}
elif isinstance(UpperCAmelCase , (np.number, np.ndarray) ) and np.issubdtype(value.dtype , np.floating ):
A_ = {"dtype": jnp.floataa}
elif config.PIL_AVAILABLE and "PIL" in sys.modules:
import PIL.Image
if isinstance(UpperCAmelCase , PIL.Image.Image ):
A_ = np.asarray(UpperCAmelCase )
# using global variable since `jaxlib.xla_extension.Device` is not serializable neither
# with `pickle` nor with `dill`, so we need to use a global variable instead
global DEVICE_MAPPING
if DEVICE_MAPPING is None:
A_ = self._map_devices_to_str()
with jax.default_device(DEVICE_MAPPING[self.device] ):
# calling jnp.array on a np.ndarray does copy the data
# see https://github.com/google/jax/issues/4486
return jnp.array(UpperCAmelCase , **{**default_dtype, **self.jnp_array_kwargs} )
def __A ( self : Any , UpperCAmelCase : Dict ):
import jax
# support for torch, tf, jax etc.
if config.TORCH_AVAILABLE and "torch" in sys.modules:
import torch
if isinstance(UpperCAmelCase , torch.Tensor ):
return self._tensorize(data_struct.detach().cpu().numpy()[()] )
if hasattr(UpperCAmelCase , "__array__" ) and not isinstance(UpperCAmelCase , jax.Array ):
A_ = data_struct.__array__()
# support for nested types like struct of list of struct
if isinstance(UpperCAmelCase , np.ndarray ):
if data_struct.dtype == object: # jax arrays cannot be instantied from an array of objects
return self._consolidate([self.recursive_tensorize(UpperCAmelCase ) for substruct in data_struct] )
elif isinstance(UpperCAmelCase , (list, tuple) ):
return self._consolidate([self.recursive_tensorize(UpperCAmelCase ) for substruct in data_struct] )
return self._tensorize(UpperCAmelCase )
def __A ( self : Tuple , UpperCAmelCase : dict ):
return map_nested(self._recursive_tensorize , UpperCAmelCase , map_list=UpperCAmelCase )
def __A ( self : Dict , UpperCAmelCase : pa.Table ):
A_ = self.numpy_arrow_extractor().extract_row(UpperCAmelCase )
A_ = self.python_features_decoder.decode_row(UpperCAmelCase )
return self.recursive_tensorize(UpperCAmelCase )
def __A ( self : Any , UpperCAmelCase : pa.Table ):
A_ = self.numpy_arrow_extractor().extract_column(UpperCAmelCase )
A_ = self.python_features_decoder.decode_column(UpperCAmelCase , pa_table.column_names[0] )
A_ = self.recursive_tensorize(UpperCAmelCase )
A_ = self._consolidate(UpperCAmelCase )
return column
def __A ( self : Dict , UpperCAmelCase : pa.Table ):
A_ = self.numpy_arrow_extractor().extract_batch(UpperCAmelCase )
A_ = self.python_features_decoder.decode_batch(UpperCAmelCase )
A_ = self.recursive_tensorize(UpperCAmelCase )
for column_name in batch:
A_ = self._consolidate(batch[column_name] )
return batch | 86 | 1 |
from math import ceil
def __snake_case ( __UpperCamelCase : int = 1001 ):
"""simple docstring"""
A_ = 1
for i in range(1 ,int(ceil(n / 2.0 ) ) ):
A_ = 2 * i + 1
A_ = 2 * i
A_ = total + 4 * odd**2 - 6 * even
return total
if __name__ == "__main__":
import sys
if len(sys.argv) == 1:
print(solution())
else:
try:
__a :Union[str, Any] = int(sys.argv[1])
print(solution(n))
except ValueError:
print('Invalid entry - please enter a number') | 86 |
import logging
import os
from typing import List, Tuple
import numpy as np
import psutil
import torch
import torch.distributed as dist
from transformers import RagRetriever
__a :Any = logging.getLogger(__name__)
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Dict , UpperCAmelCase : List[Any] , UpperCAmelCase : Optional[int] , UpperCAmelCase : Optional[Any] , UpperCAmelCase : str=None ):
super().__init__(
UpperCAmelCase , question_encoder_tokenizer=UpperCAmelCase , generator_tokenizer=UpperCAmelCase , index=UpperCAmelCase , init_retrieval=UpperCAmelCase , )
A_ = None
def __A ( self : Dict , UpperCAmelCase : int ):
logger.info("initializing retrieval" )
# initializing a separate process group for retrieval as the default
# nccl backend doesn't support gather/scatter operations while gloo
# is too slow to replace nccl for the core gpu communication
if dist.is_initialized():
logger.info("dist initialized" )
# needs to be set manually
A_ = self._infer_socket_ifname()
# avoid clash with the NCCL port
A_ = str(distributed_port + 1 )
A_ = dist.new_group(ranks=UpperCAmelCase , backend="gloo" )
# initialize retriever only on the main worker
if not dist.is_initialized() or self._is_main():
logger.info("dist not initialized / main" )
self.index.init_index()
# all processes wait untill the retriever is initialized by the main process
if dist.is_initialized():
torch.distributed.barrier(group=self.process_group )
def __A ( self : List[str] ):
return dist.get_rank(group=self.process_group ) == 0
def __A ( self : Any , UpperCAmelCase : List[Any] , UpperCAmelCase : Union[str, Any] , UpperCAmelCase : Dict=torch.floataa ):
A_ = torch.empty(UpperCAmelCase , dtype=UpperCAmelCase )
dist.scatter(UpperCAmelCase , src=0 , scatter_list=UpperCAmelCase , group=self.process_group )
return target_tensor
def __A ( self : Any ):
A_ = psutil.net_if_addrs()
# a hacky way to deal with varying network interface names
A_ = next((addr for addr in addrs if addr.startswith("e" )) , UpperCAmelCase )
return ifname
def __A ( self : Tuple , UpperCAmelCase : np.ndarray , UpperCAmelCase : int ):
# single GPU training
if not dist.is_initialized():
A_ , A_ = self._main_retrieve(UpperCAmelCase , UpperCAmelCase )
return retrieved_doc_embeds, doc_ids, self.index.get_doc_dicts(UpperCAmelCase )
# distributed training
A_ = dist.get_world_size(group=self.process_group )
# gather logic
A_ = None
if self._is_main():
A_ = [torch.empty(question_hidden_states.shape , dtype=torch.floataa ) for _ in range(UpperCAmelCase )]
dist.gather(torch.tensor(UpperCAmelCase ) , dst=0 , gather_list=UpperCAmelCase , group=self.process_group )
# scatter logic
A_ = question_hidden_states.shape[0]
A_ = []
A_ = []
if self._is_main():
assert len(UpperCAmelCase ) == world_size
A_ , A_ = self._main_retrieve(torch.cat(UpperCAmelCase ).numpy() , UpperCAmelCase )
A_ , A_ = torch.tensor(UpperCAmelCase ), torch.tensor(UpperCAmelCase )
A_ = self._chunk_tensor(UpperCAmelCase , UpperCAmelCase )
A_ = self._chunk_tensor(UpperCAmelCase , UpperCAmelCase )
A_ = self._scattered(UpperCAmelCase , [n_queries, n_docs] , target_type=torch.intaa )
A_ = self._scattered(UpperCAmelCase , [n_queries, n_docs, question_hidden_states.shape[1]] )
return retrieved_doc_embeds.numpy(), doc_ids.numpy(), self.index.get_doc_dicts(UpperCAmelCase ) | 86 | 1 |
def __snake_case ( __UpperCamelCase : int = 6008_5147_5143 ):
"""simple docstring"""
try:
A_ = int(__UpperCamelCase )
except (TypeError, ValueError):
raise TypeError("Parameter n must be int or castable to int." )
if n <= 0:
raise ValueError("Parameter n must be greater than or equal to one." )
A_ = 2
A_ = 0
if n == 2:
return 2
while n > 2:
while n % i != 0:
i += 1
A_ = i
while n % i == 0:
A_ = n // i
i += 1
return int(__UpperCamelCase )
if __name__ == "__main__":
print(F"{solution() = }") | 86 |
from jiwer import compute_measures
import datasets
__a :List[Any] = '\\n@inproceedings{inproceedings,\n author = {Morris, Andrew and Maier, Viktoria and Green, Phil},\n year = {2004},\n month = {01},\n pages = {},\n title = {From WER and RIL to MER and WIL: improved evaluation measures for connected speech recognition.}\n}\n'
__a :Union[str, Any] = '\\nWord error rate (WER) is a common metric of the performance of an automatic speech recognition system.\n\nThe general difficulty of measuring performance lies in the fact that the recognized word sequence can have a different length from the reference word sequence (supposedly the correct one). The WER is derived from the Levenshtein distance, working at the word level instead of the phoneme level. The WER is a valuable tool for comparing different systems as well as for evaluating improvements within one system. This kind of measurement, however, provides no details on the nature of translation errors and further work is therefore required to identify the main source(s) of error and to focus any research effort.\n\nThis problem is solved by first aligning the recognized word sequence with the reference (spoken) word sequence using dynamic string alignment. Examination of this issue is seen through a theory called the power law that states the correlation between perplexity and word error rate.\n\nWord error rate can then be computed as:\n\nWER = (S + D + I) / N = (S + D + I) / (S + D + C)\n\nwhere\n\nS is the number of substitutions,\nD is the number of deletions,\nI is the number of insertions,\nC is the number of correct words,\nN is the number of words in the reference (N=S+D+C).\n\nThis value indicates the average number of errors per reference word. The lower the value, the better the\nperformance of the ASR system with a WER of 0 being a perfect score.\n'
__a :str = '\nCompute WER score of transcribed segments against references.\n\nArgs:\n references: List of references for each speech input.\n predictions: List of transcriptions to score.\n concatenate_texts (bool, default=False): Whether to concatenate all input texts or compute WER iteratively.\n\nReturns:\n (float): the word error rate\n\nExamples:\n\n >>> predictions = ["this is the prediction", "there is an other sample"]\n >>> references = ["this is the reference", "there is another one"]\n >>> wer = datasets.load_metric("wer")\n >>> wer_score = wer.compute(predictions=predictions, references=references)\n >>> print(wer_score)\n 0.5\n'
@datasets.utils.file_utils.add_start_docstrings(_DESCRIPTION , _KWARGS_DESCRIPTION )
class _a ( datasets.Metric ):
"""simple docstring"""
def __A ( self : Any ):
return datasets.MetricInfo(
description=_DESCRIPTION , citation=_CITATION , inputs_description=_KWARGS_DESCRIPTION , features=datasets.Features(
{
"predictions": datasets.Value("string" , id="sequence" ),
"references": datasets.Value("string" , id="sequence" ),
} ) , codebase_urls=["https://github.com/jitsi/jiwer/"] , reference_urls=[
"https://en.wikipedia.org/wiki/Word_error_rate",
] , )
def __A ( self : Dict , UpperCAmelCase : Dict=None , UpperCAmelCase : List[Any]=None , UpperCAmelCase : str=False ):
if concatenate_texts:
return compute_measures(UpperCAmelCase , UpperCAmelCase )["wer"]
else:
A_ = 0
A_ = 0
for prediction, reference in zip(UpperCAmelCase , UpperCAmelCase ):
A_ = compute_measures(UpperCAmelCase , UpperCAmelCase )
incorrect += measures["substitutions"] + measures["deletions"] + measures["insertions"]
total += measures["substitutions"] + measures["deletions"] + measures["hits"]
return incorrect / total | 86 | 1 |
from math import factorial
def __snake_case ( __UpperCamelCase : int ,__UpperCamelCase : int ,__UpperCamelCase : float ):
"""simple docstring"""
if successes > trials:
raise ValueError("successes must be lower or equal to trials" )
if trials < 0 or successes < 0:
raise ValueError("the function is defined for non-negative integers" )
if not isinstance(__UpperCamelCase ,__UpperCamelCase ) or not isinstance(__UpperCamelCase ,__UpperCamelCase ):
raise ValueError("the function is defined for non-negative integers" )
if not 0 < prob < 1:
raise ValueError("prob has to be in range of 1 - 0" )
A_ = (prob**successes) * ((1 - prob) ** (trials - successes))
# Calculate the binomial coefficient: n! / k!(n-k)!
A_ = float(factorial(__UpperCamelCase ) )
coefficient /= factorial(__UpperCamelCase ) * factorial(trials - successes )
return probability * coefficient
if __name__ == "__main__":
from doctest import testmod
testmod()
print('Probability of 2 successes out of 4 trails')
print('with probability of 0.75 is:', end=' ')
print(binomial_distribution(2, 4, 0.75)) | 86 |
class _a :
"""simple docstring"""
def __init__( self : Union[str, Any] , UpperCAmelCase : int , UpperCAmelCase : Any , UpperCAmelCase : Dict ):
A_ = None
A_ = None
A_ = graph
self._normalize_graph(UpperCAmelCase , UpperCAmelCase )
A_ = len(UpperCAmelCase )
A_ = None
def __A ( self : Union[str, Any] , UpperCAmelCase : List[str] , UpperCAmelCase : Tuple ):
if sources is int:
A_ = [sources]
if sinks is int:
A_ = [sinks]
if len(UpperCAmelCase ) == 0 or len(UpperCAmelCase ) == 0:
return
A_ = sources[0]
A_ = sinks[0]
# make fake vertex if there are more
# than one source or sink
if len(UpperCAmelCase ) > 1 or len(UpperCAmelCase ) > 1:
A_ = 0
for i in sources:
max_input_flow += sum(self.graph[i] )
A_ = len(self.graph ) + 1
for room in self.graph:
room.insert(0 , 0 )
self.graph.insert(0 , [0] * size )
for i in sources:
A_ = max_input_flow
A_ = 0
A_ = len(self.graph ) + 1
for room in self.graph:
room.append(0 )
self.graph.append([0] * size )
for i in sinks:
A_ = max_input_flow
A_ = size - 1
def __A ( self : str ):
if self.maximum_flow_algorithm is None:
raise Exception("You need to set maximum flow algorithm before." )
if self.source_index is None or self.sink_index is None:
return 0
self.maximum_flow_algorithm.execute()
return self.maximum_flow_algorithm.getMaximumFlow()
def __A ( self : Tuple , UpperCAmelCase : List[Any] ):
A_ = algorithm(self )
class _a :
"""simple docstring"""
def __init__( self : Tuple , UpperCAmelCase : List[str] ):
A_ = flow_network
A_ = flow_network.verticesCount
A_ = flow_network.sourceIndex
A_ = flow_network.sinkIndex
# it's just a reference, so you shouldn't change
# it in your algorithms, use deep copy before doing that
A_ = flow_network.graph
A_ = False
def __A ( self : Optional[int] ):
if not self.executed:
self._algorithm()
A_ = True
def __A ( self : Dict ):
pass
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Optional[Any] , UpperCAmelCase : List[Any] ):
super().__init__(UpperCAmelCase )
# use this to save your result
A_ = -1
def __A ( self : Tuple ):
if not self.executed:
raise Exception("You should execute algorithm before using its result!" )
return self.maximum_flow
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Tuple , UpperCAmelCase : Union[str, Any] ):
super().__init__(UpperCAmelCase )
A_ = [[0] * self.verticies_count for i in range(self.verticies_count )]
A_ = [0] * self.verticies_count
A_ = [0] * self.verticies_count
def __A ( self : List[str] ):
A_ = self.verticies_count
# push some substance to graph
for nextvertex_index, bandwidth in enumerate(self.graph[self.source_index] ):
self.preflow[self.source_index][nextvertex_index] += bandwidth
self.preflow[nextvertex_index][self.source_index] -= bandwidth
self.excesses[nextvertex_index] += bandwidth
# Relabel-to-front selection rule
A_ = [
i
for i in range(self.verticies_count )
if i != self.source_index and i != self.sink_index
]
# move through list
A_ = 0
while i < len(UpperCAmelCase ):
A_ = vertices_list[i]
A_ = self.heights[vertex_index]
self.process_vertex(UpperCAmelCase )
if self.heights[vertex_index] > previous_height:
# if it was relabeled, swap elements
# and start from 0 index
vertices_list.insert(0 , vertices_list.pop(UpperCAmelCase ) )
A_ = 0
else:
i += 1
A_ = sum(self.preflow[self.source_index] )
def __A ( self : List[str] , UpperCAmelCase : Dict ):
while self.excesses[vertex_index] > 0:
for neighbour_index in range(self.verticies_count ):
# if it's neighbour and current vertex is higher
if (
self.graph[vertex_index][neighbour_index]
- self.preflow[vertex_index][neighbour_index]
> 0
and self.heights[vertex_index] > self.heights[neighbour_index]
):
self.push(UpperCAmelCase , UpperCAmelCase )
self.relabel(UpperCAmelCase )
def __A ( self : Optional[Any] , UpperCAmelCase : Dict , UpperCAmelCase : str ):
A_ = min(
self.excesses[from_index] , self.graph[from_index][to_index] - self.preflow[from_index][to_index] , )
self.preflow[from_index][to_index] += preflow_delta
self.preflow[to_index][from_index] -= preflow_delta
self.excesses[from_index] -= preflow_delta
self.excesses[to_index] += preflow_delta
def __A ( self : Optional[Any] , UpperCAmelCase : List[Any] ):
A_ = None
for to_index in range(self.verticies_count ):
if (
self.graph[vertex_index][to_index]
- self.preflow[vertex_index][to_index]
> 0
) and (min_height is None or self.heights[to_index] < min_height):
A_ = self.heights[to_index]
if min_height is not None:
A_ = min_height + 1
if __name__ == "__main__":
__a :Tuple = [0]
__a :Tuple = [3]
# graph = [
# [0, 0, 4, 6, 0, 0],
# [0, 0, 5, 2, 0, 0],
# [0, 0, 0, 0, 4, 4],
# [0, 0, 0, 0, 6, 6],
# [0, 0, 0, 0, 0, 0],
# [0, 0, 0, 0, 0, 0],
# ]
__a :List[str] = [[0, 7, 0, 0], [0, 0, 6, 0], [0, 0, 0, 8], [9, 0, 0, 0]]
# prepare our network
__a :List[str] = FlowNetwork(graph, entrances, exits)
# set algorithm
flow_network.set_maximum_flow_algorithm(PushRelabelExecutor)
# and calculate
__a :List[Any] = flow_network.find_maximum_flow()
print(F"maximum flow is {maximum_flow}") | 86 | 1 |
from .imports import is_tqdm_available
if is_tqdm_available():
from tqdm.auto import tqdm as _tqdm
from ..state import PartialState
def __snake_case ( __UpperCamelCase : bool = True ,*__UpperCamelCase : Tuple ,**__UpperCamelCase : Dict ):
"""simple docstring"""
if not is_tqdm_available():
raise ImportError("Accelerate's `tqdm` module requires `tqdm` to be installed. Please run `pip install tqdm`." )
A_ = False
if main_process_only:
A_ = PartialState().local_process_index == 0
return _tqdm(*__UpperCamelCase ,**__UpperCamelCase ,disable=__UpperCamelCase ) | 86 |
from typing import TYPE_CHECKING
from ...utils import (
OptionalDependencyNotAvailable,
_LazyModule,
is_flax_available,
is_sentencepiece_available,
is_tf_available,
is_tokenizers_available,
is_torch_available,
)
__a :Dict = {'configuration_xglm': ['XGLM_PRETRAINED_CONFIG_ARCHIVE_MAP', 'XGLMConfig']}
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Dict = ['XGLMTokenizer']
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :str = ['XGLMTokenizerFast']
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Tuple = [
'XGLM_PRETRAINED_MODEL_ARCHIVE_LIST',
'XGLMForCausalLM',
'XGLMModel',
'XGLMPreTrainedModel',
]
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :List[Any] = [
'FlaxXGLMForCausalLM',
'FlaxXGLMModel',
'FlaxXGLMPreTrainedModel',
]
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Any = [
'TF_XGLM_PRETRAINED_MODEL_ARCHIVE_LIST',
'TFXGLMForCausalLM',
'TFXGLMModel',
'TFXGLMPreTrainedModel',
]
if TYPE_CHECKING:
from .configuration_xglm import XGLM_PRETRAINED_CONFIG_ARCHIVE_MAP, XGLMConfig
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_xglm import XGLMTokenizer
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_xglm_fast import XGLMTokenizerFast
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_xglm import XGLM_PRETRAINED_MODEL_ARCHIVE_LIST, XGLMForCausalLM, XGLMModel, XGLMPreTrainedModel
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_flax_xglm import FlaxXGLMForCausalLM, FlaxXGLMModel, FlaxXGLMPreTrainedModel
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tf_xglm import (
TF_XGLM_PRETRAINED_MODEL_ARCHIVE_LIST,
TFXGLMForCausalLM,
TFXGLMModel,
TFXGLMPreTrainedModel,
)
else:
import sys
__a :List[Any] = _LazyModule(__name__, globals()['__file__'], _import_structure) | 86 | 1 |
from typing import Any, Dict, List, Union
from ..utils import add_end_docstrings, is_torch_available, is_vision_available, logging, requires_backends
from .base import PIPELINE_INIT_ARGS, Pipeline
if is_vision_available():
from ..image_utils import load_image
if is_torch_available():
import torch
from ..models.auto.modeling_auto import MODEL_FOR_OBJECT_DETECTION_MAPPING, MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING
__a :str = logging.get_logger(__name__)
__a :Any = Dict[str, Any]
__a :int = List[Prediction]
@add_end_docstrings(snake_case_ )
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Tuple , *UpperCAmelCase : Optional[int] , **UpperCAmelCase : Optional[Any] ):
super().__init__(*UpperCAmelCase , **UpperCAmelCase )
if self.framework == "tf":
raise ValueError(f'''The {self.__class__} is only available in PyTorch.''' )
requires_backends(self , "vision" )
self.check_model_type(
dict(MODEL_FOR_OBJECT_DETECTION_MAPPING.items() + MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING.items() ) )
def __A ( self : str , **UpperCAmelCase : str ):
A_ = {}
if "threshold" in kwargs:
A_ = kwargs["threshold"]
return {}, {}, postprocess_kwargs
def __call__( self : Union[str, Any] , *UpperCAmelCase : Union[str, Any] , **UpperCAmelCase : Optional[Any] ):
return super().__call__(*UpperCAmelCase , **UpperCAmelCase )
def __A ( self : str , UpperCAmelCase : Any ):
A_ = load_image(UpperCAmelCase )
A_ = torch.IntTensor([[image.height, image.width]] )
A_ = self.image_processor(images=[image] , return_tensors="pt" )
if self.tokenizer is not None:
A_ = self.tokenizer(text=inputs["words"] , boxes=inputs["boxes"] , return_tensors="pt" )
A_ = target_size
return inputs
def __A ( self : Optional[Any] , UpperCAmelCase : Optional[int] ):
A_ = model_inputs.pop("target_size" )
A_ = self.model(**UpperCAmelCase )
A_ = outputs.__class__({"target_size": target_size, **outputs} )
if self.tokenizer is not None:
A_ = model_inputs["bbox"]
return model_outputs
def __A ( self : str , UpperCAmelCase : Optional[int] , UpperCAmelCase : Union[str, Any]=0.9 ):
A_ = model_outputs["target_size"]
if self.tokenizer is not None:
# This is a LayoutLMForTokenClassification variant.
# The OCR got the boxes and the model classified the words.
A_ , A_ = target_size[0].tolist()
def unnormalize(UpperCAmelCase : Any ):
return self._get_bounding_box(
torch.Tensor(
[
(width * bbox[0] / 1000),
(height * bbox[1] / 1000),
(width * bbox[2] / 1000),
(height * bbox[3] / 1000),
] ) )
A_ , A_ = model_outputs["logits"].squeeze(0 ).softmax(dim=-1 ).max(dim=-1 )
A_ = [self.model.config.idalabel[prediction] for prediction in classes.tolist()]
A_ = [unnormalize(UpperCAmelCase ) for bbox in model_outputs["bbox"].squeeze(0 )]
A_ = ["score", "label", "box"]
A_ = [dict(zip(UpperCAmelCase , UpperCAmelCase ) ) for vals in zip(scores.tolist() , UpperCAmelCase , UpperCAmelCase ) if vals[0] > threshold]
else:
# This is a regular ForObjectDetectionModel
A_ = self.image_processor.post_process_object_detection(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase )
A_ = raw_annotations[0]
A_ = raw_annotation["scores"]
A_ = raw_annotation["labels"]
A_ = raw_annotation["boxes"]
A_ = scores.tolist()
A_ = [self.model.config.idalabel[label.item()] for label in labels]
A_ = [self._get_bounding_box(UpperCAmelCase ) for box in boxes]
# {"scores": [...], ...} --> [{"score":x, ...}, ...]
A_ = ["score", "label", "box"]
A_ = [
dict(zip(UpperCAmelCase , UpperCAmelCase ) )
for vals in zip(raw_annotation["scores"] , raw_annotation["labels"] , raw_annotation["boxes"] )
]
return annotation
def __A ( self : Tuple , UpperCAmelCase : "torch.Tensor" ):
if self.framework != "pt":
raise ValueError("The ObjectDetectionPipeline is only available in PyTorch." )
A_ , A_ , A_ , A_ = box.int().tolist()
A_ = {
"xmin": xmin,
"ymin": ymin,
"xmax": xmax,
"ymax": ymax,
}
return bbox | 86 |
# Usage:
# ./gen-card-facebook-wmt19.py
import os
from pathlib import Path
def __snake_case ( __UpperCamelCase : str ,__UpperCamelCase : str ,__UpperCamelCase : List[str] ):
"""simple docstring"""
A_ = {
"en": "Machine learning is great, isn't it?",
"ru": "Машинное обучение - это здорово, не так ли?",
"de": "Maschinelles Lernen ist großartig, oder?",
}
# BLUE scores as follows:
# "pair": [fairseq, transformers]
A_ = {
"ru-en": ["[41.3](http://matrix.statmt.org/matrix/output/1907?run_id=6937)", "39.20"],
"en-ru": ["[36.4](http://matrix.statmt.org/matrix/output/1914?run_id=6724)", "33.47"],
"en-de": ["[43.1](http://matrix.statmt.org/matrix/output/1909?run_id=6862)", "42.83"],
"de-en": ["[42.3](http://matrix.statmt.org/matrix/output/1902?run_id=6750)", "41.35"],
}
A_ = f'''{src_lang}-{tgt_lang}'''
A_ = f'''
---
language:
- {src_lang}
- {tgt_lang}
thumbnail:
tags:
- translation
- wmt19
- facebook
license: apache-2.0
datasets:
- wmt19
metrics:
- bleu
---
# FSMT
## Model description
This is a ported version of [fairseq wmt19 transformer](https://github.com/pytorch/fairseq/blob/master/examples/wmt19/README.md) for {src_lang}-{tgt_lang}.
For more details, please see, [Facebook FAIR\'s WMT19 News Translation Task Submission](https://arxiv.org/abs/1907.06616).
The abbreviation FSMT stands for FairSeqMachineTranslation
All four models are available:
* [wmt19-en-ru](https://huggingface.co/facebook/wmt19-en-ru)
* [wmt19-ru-en](https://huggingface.co/facebook/wmt19-ru-en)
* [wmt19-en-de](https://huggingface.co/facebook/wmt19-en-de)
* [wmt19-de-en](https://huggingface.co/facebook/wmt19-de-en)
## Intended uses & limitations
#### How to use
```python
from transformers import FSMTForConditionalGeneration, FSMTTokenizer
mname = "facebook/wmt19-{src_lang}-{tgt_lang}"
tokenizer = FSMTTokenizer.from_pretrained(mname)
model = FSMTForConditionalGeneration.from_pretrained(mname)
input = "{texts[src_lang]}"
input_ids = tokenizer.encode(input, return_tensors="pt")
outputs = model.generate(input_ids)
decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(decoded) # {texts[tgt_lang]}
```
#### Limitations and bias
- The original (and this ported model) doesn\'t seem to handle well inputs with repeated sub-phrases, [content gets truncated](https://discuss.huggingface.co/t/issues-with-translating-inputs-containing-repeated-phrases/981)
## Training data
Pretrained weights were left identical to the original model released by fairseq. For more details, please, see the [paper](https://arxiv.org/abs/1907.06616).
## Eval results
pair | fairseq | transformers
-------|---------|----------
{pair} | {scores[pair][0]} | {scores[pair][1]}
The score is slightly below the score reported by `fairseq`, since `transformers`` currently doesn\'t support:
- model ensemble, therefore the best performing checkpoint was ported (``model4.pt``).
- re-ranking
The score was calculated using this code:
```bash
git clone https://github.com/huggingface/transformers
cd transformers
export PAIR={pair}
export DATA_DIR=data/$PAIR
export SAVE_DIR=data/$PAIR
export BS=8
export NUM_BEAMS=15
mkdir -p $DATA_DIR
sacrebleu -t wmt19 -l $PAIR --echo src > $DATA_DIR/val.source
sacrebleu -t wmt19 -l $PAIR --echo ref > $DATA_DIR/val.target
echo $PAIR
PYTHONPATH="src:examples/seq2seq" python examples/seq2seq/run_eval.py facebook/wmt19-$PAIR $DATA_DIR/val.source $SAVE_DIR/test_translations.txt --reference_path $DATA_DIR/val.target --score_path $SAVE_DIR/test_bleu.json --bs $BS --task translation --num_beams $NUM_BEAMS
```
note: fairseq reports using a beam of 50, so you should get a slightly higher score if re-run with `--num_beams 50`.
## Data Sources
- [training, etc.](http://www.statmt.org/wmt19/)
- [test set](http://matrix.statmt.org/test_sets/newstest2019.tgz?1556572561)
### BibTeX entry and citation info
```bibtex
@inproceedings{{...,
year={{2020}},
title={{Facebook FAIR\'s WMT19 News Translation Task Submission}},
author={{Ng, Nathan and Yee, Kyra and Baevski, Alexei and Ott, Myle and Auli, Michael and Edunov, Sergey}},
booktitle={{Proc. of WMT}},
}}
```
## TODO
- port model ensemble (fairseq uses 4 model checkpoints)
'''
os.makedirs(__UpperCamelCase ,exist_ok=__UpperCamelCase )
A_ = os.path.join(__UpperCamelCase ,"README.md" )
print(f'''Generating {path}''' )
with open(__UpperCamelCase ,"w" ,encoding="utf-8" ) as f:
f.write(__UpperCamelCase )
# make sure we are under the root of the project
__a :Optional[Any] = Path(__file__).resolve().parent.parent.parent
__a :Optional[Any] = repo_dir / 'model_cards'
for model_name in ["wmt19-ru-en", "wmt19-en-ru", "wmt19-en-de", "wmt19-de-en"]:
__a , __a , __a :int = model_name.split('-')
__a :str = model_cards_dir / 'facebook' / model_name
write_model_card(model_card_dir, src_lang=src_lang, tgt_lang=tgt_lang) | 86 | 1 |
import numpy as np
class _a :
"""simple docstring"""
def __init__( self : str ):
A_ = (0, 0)
A_ = None
A_ = 0
A_ = 0
A_ = 0
def __eq__( self : str , UpperCAmelCase : str ):
return self.position == cell.position
def __A ( self : Union[str, Any] ):
print(self.position )
class _a :
"""simple docstring"""
def __init__( self : Tuple , UpperCAmelCase : Union[str, Any]=(5, 5) ):
A_ = np.zeros(UpperCAmelCase )
A_ = world_size[0]
A_ = world_size[1]
def __A ( self : Dict ):
print(self.w )
def __A ( self : List[Any] , UpperCAmelCase : List[str] ):
A_ = [
(-1, -1),
(-1, 0),
(-1, 1),
(0, -1),
(0, 1),
(1, -1),
(1, 0),
(1, 1),
]
A_ = cell.position[0]
A_ = cell.position[1]
A_ = []
for n in neughbour_cord:
A_ = current_x + n[0]
A_ = current_y + n[1]
if 0 <= x < self.world_x_limit and 0 <= y < self.world_y_limit:
A_ = Cell()
A_ = (x, y)
A_ = cell
neighbours.append(UpperCAmelCase )
return neighbours
def __snake_case ( __UpperCamelCase : Any ,__UpperCamelCase : str ,__UpperCamelCase : List[Any] ):
"""simple docstring"""
A_ = []
A_ = []
_open.append(__UpperCamelCase )
while _open:
A_ = np.argmin([n.f for n in _open] )
A_ = _open[min_f]
_closed.append(_open.pop(__UpperCamelCase ) )
if current == goal:
break
for n in world.get_neigbours(__UpperCamelCase ):
for c in _closed:
if c == n:
continue
A_ = current.g + 1
A_ , A_ = n.position
A_ , A_ = goal.position
A_ = (ya - ya) ** 2 + (xa - xa) ** 2
A_ = n.h + n.g
for c in _open:
if c == n and c.f < n.f:
continue
_open.append(__UpperCamelCase )
A_ = []
while current.parent is not None:
path.append(current.position )
A_ = current.parent
path.append(current.position )
return path[::-1]
if __name__ == "__main__":
__a :List[Any] = Gridworld()
# Start position and goal
__a :str = Cell()
__a :Union[str, Any] = (0, 0)
__a :int = Cell()
__a :int = (4, 4)
print(F"path from {start.position} to {goal.position}")
__a :List[str] = astar(world, start, goal)
# Just for visual reasons.
for i in s:
__a :Tuple = 1
print(world.w) | 86 |
from ..utils import DummyObject, requires_backends
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Optional[Any] = ['torch', 'transformers', 'onnx']
def __init__( self : str , *UpperCAmelCase : int , **UpperCAmelCase : List[Any] ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Dict , *UpperCAmelCase : Dict , **UpperCAmelCase : Union[str, Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[Any] , *UpperCAmelCase : List[Any] , **UpperCAmelCase : Union[str, Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : str = ['torch', 'transformers', 'onnx']
def __init__( self : Optional[int] , *UpperCAmelCase : List[str] , **UpperCAmelCase : int ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : List[str] , *UpperCAmelCase : List[str] , **UpperCAmelCase : List[Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[Any] , *UpperCAmelCase : Dict , **UpperCAmelCase : Optional[Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Optional[Any] = ['torch', 'transformers', 'onnx']
def __init__( self : Union[str, Any] , *UpperCAmelCase : Any , **UpperCAmelCase : Dict ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Tuple , *UpperCAmelCase : Dict , **UpperCAmelCase : str ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Dict , *UpperCAmelCase : List[str] , **UpperCAmelCase : List[str] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : int = ['torch', 'transformers', 'onnx']
def __init__( self : List[str] , *UpperCAmelCase : Dict , **UpperCAmelCase : Optional[Any] ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Any , *UpperCAmelCase : Union[str, Any] , **UpperCAmelCase : Dict ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[Any] , *UpperCAmelCase : int , **UpperCAmelCase : Optional[int] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Dict = ['torch', 'transformers', 'onnx']
def __init__( self : List[str] , *UpperCAmelCase : str , **UpperCAmelCase : int ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : List[str] , *UpperCAmelCase : Optional[int] , **UpperCAmelCase : Optional[int] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[int] , *UpperCAmelCase : int , **UpperCAmelCase : List[Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : int = ['torch', 'transformers', 'onnx']
def __init__( self : Tuple , *UpperCAmelCase : List[Any] , **UpperCAmelCase : Optional[Any] ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[int] , *UpperCAmelCase : Dict , **UpperCAmelCase : str ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[int] , *UpperCAmelCase : Optional[Any] , **UpperCAmelCase : int ):
requires_backends(cls , ["torch", "transformers", "onnx"] ) | 86 | 1 |
def __snake_case ( __UpperCamelCase : list ,__UpperCamelCase : list ,__UpperCamelCase : int ):
"""simple docstring"""
A_ = len(__UpperCamelCase )
A_ = [[0] * n for i in range(__UpperCamelCase )]
for i in range(__UpperCamelCase ):
A_ = y_points[i]
for i in range(2 ,__UpperCamelCase ):
for j in range(__UpperCamelCase ,__UpperCamelCase ):
A_ = (
(xa - x_points[j - i + 1]) * q[j][i - 1]
- (xa - x_points[j]) * q[j - 1][i - 1]
) / (x_points[j] - x_points[j - i + 1])
return [q[n - 1][n - 1], q]
if __name__ == "__main__":
import doctest
doctest.testmod() | 86 |
import torch
from diffusers import DDPMParallelScheduler
from .test_schedulers import SchedulerCommonTest
class _a ( snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Optional[Any] = (DDPMParallelScheduler,)
def __A ( self : List[Any] , **UpperCAmelCase : Optional[int] ):
A_ = {
"num_train_timesteps": 1000,
"beta_start": 0.0_001,
"beta_end": 0.02,
"beta_schedule": "linear",
"variance_type": "fixed_small",
"clip_sample": True,
}
config.update(**UpperCAmelCase )
return config
def __A ( self : Optional[Any] ):
for timesteps in [1, 5, 100, 1000]:
self.check_over_configs(num_train_timesteps=UpperCAmelCase )
def __A ( self : Dict ):
for beta_start, beta_end in zip([0.0_001, 0.001, 0.01, 0.1] , [0.002, 0.02, 0.2, 2] ):
self.check_over_configs(beta_start=UpperCAmelCase , beta_end=UpperCAmelCase )
def __A ( self : int ):
for schedule in ["linear", "squaredcos_cap_v2"]:
self.check_over_configs(beta_schedule=UpperCAmelCase )
def __A ( self : Tuple ):
for variance in ["fixed_small", "fixed_large", "other"]:
self.check_over_configs(variance_type=UpperCAmelCase )
def __A ( self : int ):
for clip_sample in [True, False]:
self.check_over_configs(clip_sample=UpperCAmelCase )
def __A ( self : Union[str, Any] ):
self.check_over_configs(thresholding=UpperCAmelCase )
for threshold in [0.5, 1.0, 2.0]:
for prediction_type in ["epsilon", "sample", "v_prediction"]:
self.check_over_configs(
thresholding=UpperCAmelCase , prediction_type=UpperCAmelCase , sample_max_value=UpperCAmelCase , )
def __A ( self : Optional[int] ):
for prediction_type in ["epsilon", "sample", "v_prediction"]:
self.check_over_configs(prediction_type=UpperCAmelCase )
def __A ( self : Tuple ):
for t in [0, 500, 999]:
self.check_over_forward(time_step=UpperCAmelCase )
def __A ( self : Tuple ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
assert torch.sum(torch.abs(scheduler._get_variance(0 ) - 0.0 ) ) < 1E-5
assert torch.sum(torch.abs(scheduler._get_variance(487 ) - 0.00_979 ) ) < 1E-5
assert torch.sum(torch.abs(scheduler._get_variance(999 ) - 0.02 ) ) < 1E-5
def __A ( self : List[Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = len(UpperCAmelCase )
A_ = self.dummy_model()
A_ = self.dummy_sample_deter
A_ = self.dummy_sample_deter + 0.1
A_ = self.dummy_sample_deter - 0.1
A_ = samplea.shape[0]
A_ = torch.stack([samplea, samplea, samplea] , dim=0 )
A_ = torch.arange(UpperCAmelCase )[0:3, None].repeat(1 , UpperCAmelCase )
A_ = model(samples.flatten(0 , 1 ) , timesteps.flatten(0 , 1 ) )
A_ = scheduler.batch_step_no_noise(UpperCAmelCase , timesteps.flatten(0 , 1 ) , samples.flatten(0 , 1 ) )
A_ = torch.sum(torch.abs(UpperCAmelCase ) )
A_ = torch.mean(torch.abs(UpperCAmelCase ) )
assert abs(result_sum.item() - 1_153.1_833 ) < 1E-2
assert abs(result_mean.item() - 0.5_005 ) < 1E-3
def __A ( self : Tuple ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = len(UpperCAmelCase )
A_ = self.dummy_model()
A_ = self.dummy_sample_deter
A_ = torch.manual_seed(0 )
for t in reversed(range(UpperCAmelCase ) ):
# 1. predict noise residual
A_ = model(UpperCAmelCase , UpperCAmelCase )
# 2. predict previous mean of sample x_t-1
A_ = scheduler.step(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase , generator=UpperCAmelCase ).prev_sample
A_ = pred_prev_sample
A_ = torch.sum(torch.abs(UpperCAmelCase ) )
A_ = torch.mean(torch.abs(UpperCAmelCase ) )
assert abs(result_sum.item() - 258.9_606 ) < 1E-2
assert abs(result_mean.item() - 0.3_372 ) < 1E-3
def __A ( self : Tuple ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config(prediction_type="v_prediction" )
A_ = scheduler_class(**UpperCAmelCase )
A_ = len(UpperCAmelCase )
A_ = self.dummy_model()
A_ = self.dummy_sample_deter
A_ = torch.manual_seed(0 )
for t in reversed(range(UpperCAmelCase ) ):
# 1. predict noise residual
A_ = model(UpperCAmelCase , UpperCAmelCase )
# 2. predict previous mean of sample x_t-1
A_ = scheduler.step(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase , generator=UpperCAmelCase ).prev_sample
A_ = pred_prev_sample
A_ = torch.sum(torch.abs(UpperCAmelCase ) )
A_ = torch.mean(torch.abs(UpperCAmelCase ) )
assert abs(result_sum.item() - 202.0_296 ) < 1E-2
assert abs(result_mean.item() - 0.2_631 ) < 1E-3
def __A ( self : Union[str, Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = [100, 87, 50, 1, 0]
scheduler.set_timesteps(timesteps=UpperCAmelCase )
A_ = scheduler.timesteps
for i, timestep in enumerate(UpperCAmelCase ):
if i == len(UpperCAmelCase ) - 1:
A_ = -1
else:
A_ = timesteps[i + 1]
A_ = scheduler.previous_timestep(UpperCAmelCase )
A_ = prev_t.item()
self.assertEqual(UpperCAmelCase , UpperCAmelCase )
def __A ( self : List[Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = [100, 87, 50, 51, 0]
with self.assertRaises(UpperCAmelCase , msg="`custom_timesteps` must be in descending order." ):
scheduler.set_timesteps(timesteps=UpperCAmelCase )
def __A ( self : List[Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = [100, 87, 50, 1, 0]
A_ = len(UpperCAmelCase )
with self.assertRaises(UpperCAmelCase , msg="Can only pass one of `num_inference_steps` or `custom_timesteps`." ):
scheduler.set_timesteps(num_inference_steps=UpperCAmelCase , timesteps=UpperCAmelCase )
def __A ( self : Optional[Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = [scheduler.config.num_train_timesteps]
with self.assertRaises(
UpperCAmelCase , msg="`timesteps` must start before `self.config.train_timesteps`: {scheduler.config.num_train_timesteps}}" , ):
scheduler.set_timesteps(timesteps=UpperCAmelCase ) | 86 | 1 |
from typing import TYPE_CHECKING
from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_flax_available, is_torch_available
__a :List[Any] = {'configuration_speech_encoder_decoder': ['SpeechEncoderDecoderConfig']}
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Dict = ['SpeechEncoderDecoderModel']
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :List[str] = ['FlaxSpeechEncoderDecoderModel']
if TYPE_CHECKING:
from .configuration_speech_encoder_decoder import SpeechEncoderDecoderConfig
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_speech_encoder_decoder import SpeechEncoderDecoderModel
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_flax_speech_encoder_decoder import FlaxSpeechEncoderDecoderModel
else:
import sys
__a :str = _LazyModule(__name__, globals()['__file__'], _import_structure, module_spec=__spec__) | 86 |
import argparse
import json
import os
import torch
from transformers import LukeConfig, LukeModel, LukeTokenizer, RobertaTokenizer
from transformers.tokenization_utils_base import AddedToken
@torch.no_grad()
def __snake_case ( __UpperCamelCase : Tuple ,__UpperCamelCase : Dict ,__UpperCamelCase : Union[str, Any] ,__UpperCamelCase : List[Any] ,__UpperCamelCase : List[Any] ):
"""simple docstring"""
with open(__UpperCamelCase ) as metadata_file:
A_ = json.load(__UpperCamelCase )
A_ = LukeConfig(use_entity_aware_attention=__UpperCamelCase ,**metadata["model_config"] )
# Load in the weights from the checkpoint_path
A_ = torch.load(__UpperCamelCase ,map_location="cpu" )
# Load the entity vocab file
A_ = load_entity_vocab(__UpperCamelCase )
A_ = RobertaTokenizer.from_pretrained(metadata["model_config"]["bert_model_name"] )
# Add special tokens to the token vocabulary for downstream tasks
A_ = AddedToken("<ent>" ,lstrip=__UpperCamelCase ,rstrip=__UpperCamelCase )
A_ = AddedToken("<ent2>" ,lstrip=__UpperCamelCase ,rstrip=__UpperCamelCase )
tokenizer.add_special_tokens({"additional_special_tokens": [entity_token_a, entity_token_a]} )
config.vocab_size += 2
print(f'''Saving tokenizer to {pytorch_dump_folder_path}''' )
tokenizer.save_pretrained(__UpperCamelCase )
with open(os.path.join(__UpperCamelCase ,LukeTokenizer.vocab_files_names["entity_vocab_file"] ) ,"w" ) as f:
json.dump(__UpperCamelCase ,__UpperCamelCase )
A_ = LukeTokenizer.from_pretrained(__UpperCamelCase )
# Initialize the embeddings of the special tokens
A_ = state_dict["embeddings.word_embeddings.weight"]
A_ = word_emb[tokenizer.convert_tokens_to_ids(["@"] )[0]].unsqueeze(0 )
A_ = word_emb[tokenizer.convert_tokens_to_ids(["#"] )[0]].unsqueeze(0 )
A_ = torch.cat([word_emb, ent_emb, enta_emb] )
# Initialize the query layers of the entity-aware self-attention mechanism
for layer_index in range(config.num_hidden_layers ):
for matrix_name in ["query.weight", "query.bias"]:
A_ = f'''encoder.layer.{layer_index}.attention.self.'''
A_ = state_dict[prefix + matrix_name]
A_ = state_dict[prefix + matrix_name]
A_ = state_dict[prefix + matrix_name]
# Initialize the embedding of the [MASK2] entity using that of the [MASK] entity for downstream tasks
A_ = state_dict["entity_embeddings.entity_embeddings.weight"]
A_ = entity_emb[entity_vocab["[MASK]"]]
A_ = LukeModel(config=__UpperCamelCase ).eval()
A_ , A_ = model.load_state_dict(__UpperCamelCase ,strict=__UpperCamelCase )
if not (len(__UpperCamelCase ) == 1 and missing_keys[0] == "embeddings.position_ids"):
raise ValueError(f'''Missing keys {", ".join(__UpperCamelCase )}. Expected only missing embeddings.position_ids''' )
if not (all(key.startswith("entity_predictions" ) or key.startswith("lm_head" ) for key in unexpected_keys )):
raise ValueError(
"Unexpected keys"
f''' {", ".join([key for key in unexpected_keys if not (key.startswith("entity_predictions" ) or key.startswith("lm_head" ))] )}''' )
# Check outputs
A_ = LukeTokenizer.from_pretrained(__UpperCamelCase ,task="entity_classification" )
A_ = (
"Top seed Ana Ivanovic said on Thursday she could hardly believe her luck as a fortuitous netcord helped the"
" new world number one avoid a humiliating second- round exit at Wimbledon ."
)
A_ = (39, 42)
A_ = tokenizer(__UpperCamelCase ,entity_spans=[span] ,add_prefix_space=__UpperCamelCase ,return_tensors="pt" )
A_ = model(**__UpperCamelCase )
# Verify word hidden states
if model_size == "large":
A_ = torch.Size((1, 42, 1024) )
A_ = torch.tensor(
[[0.0133, 0.0865, 0.0095], [0.3093, -0.2576, -0.7418], [-0.1720, -0.2117, -0.2869]] )
else: # base
A_ = torch.Size((1, 42, 768) )
A_ = torch.tensor([[0.0037, 0.1368, -0.0091], [0.1099, 0.3329, -0.1095], [0.0765, 0.5335, 0.1179]] )
if not (outputs.last_hidden_state.shape == expected_shape):
raise ValueError(
f'''Outputs.last_hidden_state.shape is {outputs.last_hidden_state.shape}, Expected shape is {expected_shape}''' )
if not torch.allclose(outputs.last_hidden_state[0, :3, :3] ,__UpperCamelCase ,atol=1E-4 ):
raise ValueError
# Verify entity hidden states
if model_size == "large":
A_ = torch.Size((1, 1, 1024) )
A_ = torch.tensor([[0.0466, -0.0106, -0.0179]] )
else: # base
A_ = torch.Size((1, 1, 768) )
A_ = torch.tensor([[0.1457, 0.1044, 0.0174]] )
if not (outputs.entity_last_hidden_state.shape != expected_shape):
raise ValueError(
f'''Outputs.entity_last_hidden_state.shape is {outputs.entity_last_hidden_state.shape}, Expected shape is'''
f''' {expected_shape}''' )
if not torch.allclose(outputs.entity_last_hidden_state[0, :3, :3] ,__UpperCamelCase ,atol=1E-4 ):
raise ValueError
# Finally, save our PyTorch model and tokenizer
print("Saving PyTorch model to {}".format(__UpperCamelCase ) )
model.save_pretrained(__UpperCamelCase )
def __snake_case ( __UpperCamelCase : str ):
"""simple docstring"""
A_ = {}
with open(__UpperCamelCase ,"r" ,encoding="utf-8" ) as f:
for index, line in enumerate(__UpperCamelCase ):
A_ , A_ = line.rstrip().split("\t" )
A_ = index
return entity_vocab
if __name__ == "__main__":
__a :Dict = argparse.ArgumentParser()
# Required parameters
parser.add_argument('--checkpoint_path', type=str, help='Path to a pytorch_model.bin file.')
parser.add_argument(
'--metadata_path', default=None, type=str, help='Path to a metadata.json file, defining the configuration.'
)
parser.add_argument(
'--entity_vocab_path',
default=None,
type=str,
help='Path to an entity_vocab.tsv file, containing the entity vocabulary.',
)
parser.add_argument(
'--pytorch_dump_folder_path', default=None, type=str, help='Path to where to dump the output PyTorch model.'
)
parser.add_argument(
'--model_size', default='base', type=str, choices=['base', 'large'], help='Size of the model to be converted.'
)
__a :Tuple = parser.parse_args()
convert_luke_checkpoint(
args.checkpoint_path,
args.metadata_path,
args.entity_vocab_path,
args.pytorch_dump_folder_path,
args.model_size,
) | 86 | 1 |
def __snake_case ( __UpperCamelCase : int = 1000 ):
"""simple docstring"""
A_ = 2**power
A_ = str(__UpperCamelCase )
A_ = list(__UpperCamelCase )
A_ = 0
for i in list_num:
sum_of_num += int(__UpperCamelCase )
return sum_of_num
if __name__ == "__main__":
__a :Union[str, Any] = int(input('Enter the power of 2: ').strip())
print('2 ^ ', power, ' = ', 2**power)
__a :List[str] = solution(power)
print('Sum of the digits is: ', result) | 86 |
import math
import os
from copy import deepcopy
import datasets
import evaluate
import torch
import transformers
from datasets import load_dataset
from torch.utils.data import DataLoader
from transformers import AutoModelForSequenceClassification, AutoTokenizer
from accelerate import Accelerator
from accelerate.test_utils import RegressionDataset, RegressionModel
from accelerate.utils import is_tpu_available, set_seed
__a :Optional[Any] = 'true'
def __snake_case ( __UpperCamelCase : Any ,__UpperCamelCase : List[Any]=82 ,__UpperCamelCase : Dict=16 ):
"""simple docstring"""
set_seed(42 )
A_ = RegressionModel()
A_ = deepcopy(__UpperCamelCase )
A_ = RegressionDataset(length=__UpperCamelCase )
A_ = DataLoader(__UpperCamelCase ,batch_size=__UpperCamelCase )
model.to(accelerator.device )
A_ , A_ = accelerator.prepare(__UpperCamelCase ,__UpperCamelCase )
return model, ddp_model, dataloader
def __snake_case ( __UpperCamelCase : Accelerator ,__UpperCamelCase : Dict=False ):
"""simple docstring"""
A_ = AutoTokenizer.from_pretrained("hf-internal-testing/mrpc-bert-base-cased" )
A_ = load_dataset("glue" ,"mrpc" ,split="validation" )
def tokenize_function(__UpperCamelCase : Optional[Any] ):
A_ = tokenizer(examples["sentence1"] ,examples["sentence2"] ,truncation=__UpperCamelCase ,max_length=__UpperCamelCase )
return outputs
with accelerator.main_process_first():
A_ = dataset.map(
__UpperCamelCase ,batched=__UpperCamelCase ,remove_columns=["idx", "sentence1", "sentence2"] ,)
A_ = tokenized_datasets.rename_column("label" ,"labels" )
def collate_fn(__UpperCamelCase : Union[str, Any] ):
if use_longest:
return tokenizer.pad(__UpperCamelCase ,padding="longest" ,return_tensors="pt" )
return tokenizer.pad(__UpperCamelCase ,padding="max_length" ,max_length=128 ,return_tensors="pt" )
return DataLoader(__UpperCamelCase ,shuffle=__UpperCamelCase ,collate_fn=__UpperCamelCase ,batch_size=16 )
def __snake_case ( __UpperCamelCase : Optional[Any] ,__UpperCamelCase : str ):
"""simple docstring"""
A_ = Accelerator(dispatch_batches=__UpperCamelCase ,split_batches=__UpperCamelCase )
A_ = get_dataloader(__UpperCamelCase ,not dispatch_batches )
A_ = AutoModelForSequenceClassification.from_pretrained(
"hf-internal-testing/mrpc-bert-base-cased" ,return_dict=__UpperCamelCase )
A_ , A_ = accelerator.prepare(__UpperCamelCase ,__UpperCamelCase )
return {"ddp": [ddp_model, ddp_dataloader, "cuda:0"], "no": [model, dataloader, accelerator.device]}, accelerator
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : int ,__UpperCamelCase : Optional[Any] ):
"""simple docstring"""
A_ = []
for batch in dataloader:
A_ , A_ = batch.values()
with torch.no_grad():
A_ = model(__UpperCamelCase )
A_ , A_ = accelerator.gather_for_metrics((logit, target) )
logits_and_targets.append((logit, target) )
A_ , A_ = [], []
for logit, targ in logits_and_targets:
logits.append(__UpperCamelCase )
targs.append(__UpperCamelCase )
A_ , A_ = torch.cat(__UpperCamelCase ), torch.cat(__UpperCamelCase )
return logits, targs
def __snake_case ( __UpperCamelCase : Accelerator ,__UpperCamelCase : Dict=82 ,__UpperCamelCase : List[Any]=False ,__UpperCamelCase : Dict=False ,__UpperCamelCase : Optional[int]=16 ):
"""simple docstring"""
A_ , A_ , A_ = get_basic_setup(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
A_ , A_ = generate_predictions(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
assert (
len(__UpperCamelCase ) == num_samples
), f'''Unexpected number of inputs:\n Expected: {num_samples}\n Actual: {len(__UpperCamelCase )}'''
def __snake_case ( __UpperCamelCase : bool = False ,__UpperCamelCase : bool = False ):
"""simple docstring"""
A_ = evaluate.load("glue" ,"mrpc" )
A_ , A_ = get_mrpc_setup(__UpperCamelCase ,__UpperCamelCase )
# First do baseline
A_ , A_ , A_ = setup["no"]
model.to(__UpperCamelCase )
model.eval()
for batch in dataloader:
batch.to(__UpperCamelCase )
with torch.inference_mode():
A_ = model(**__UpperCamelCase )
A_ = outputs.logits.argmax(dim=-1 )
metric.add_batch(predictions=__UpperCamelCase ,references=batch["labels"] )
A_ = metric.compute()
# Then do distributed
A_ , A_ , A_ = setup["ddp"]
model.eval()
for batch in dataloader:
with torch.inference_mode():
A_ = model(**__UpperCamelCase )
A_ = outputs.logits.argmax(dim=-1 )
A_ = batch["labels"]
A_ , A_ = accelerator.gather_for_metrics((preds, references) )
metric.add_batch(predictions=__UpperCamelCase ,references=__UpperCamelCase )
A_ = metric.compute()
for key in "accuracy f1".split():
assert math.isclose(
baseline[key] ,distributed[key] ), f'''Baseline and Distributed are not the same for key {key}:\n\tBaseline: {baseline[key]}\n\tDistributed: {distributed[key]}\n'''
def __snake_case ( ):
"""simple docstring"""
A_ = Accelerator(split_batches=__UpperCamelCase ,dispatch_batches=__UpperCamelCase )
if accelerator.is_local_main_process:
datasets.utils.logging.set_verbosity_warning()
transformers.utils.logging.set_verbosity_warning()
else:
datasets.utils.logging.set_verbosity_error()
transformers.utils.logging.set_verbosity_error()
# These are a bit slower so they should only be ran on the GPU or TPU
if torch.cuda.is_available() or is_tpu_available():
if accelerator.is_local_main_process:
print("**Testing gather_for_metrics**" )
for split_batches in [True, False]:
for dispatch_batches in [True, False]:
if accelerator.is_local_main_process:
print(f'''With: `split_batches={split_batches}`, `dispatch_batches={dispatch_batches}`''' )
test_mrpc(__UpperCamelCase ,__UpperCamelCase )
accelerator.state._reset_state()
if accelerator.is_local_main_process:
print("**Test torch metrics**" )
for split_batches in [True, False]:
for dispatch_batches in [True, False]:
A_ = Accelerator(split_batches=__UpperCamelCase ,dispatch_batches=__UpperCamelCase )
if accelerator.is_local_main_process:
print(f'''With: `split_batches={split_batches}`, `dispatch_batches={dispatch_batches}`, length=99''' )
test_torch_metrics(__UpperCamelCase ,99 )
accelerator.state._reset_state()
if accelerator.is_local_main_process:
print("**Test last batch is not dropped when perfectly divisible**" )
A_ = Accelerator()
test_torch_metrics(__UpperCamelCase ,512 )
accelerator.state._reset_state()
def __snake_case ( __UpperCamelCase : List[Any] ):
"""simple docstring"""
main()
if __name__ == "__main__":
main() | 86 | 1 |
from ...configuration_utils import PretrainedConfig
from ...utils import logging
__a :List[str] = logging.get_logger(__name__)
__a :List[str] = {
's-JoL/Open-Llama-V1': 'https://huggingface.co/s-JoL/Open-Llama-V1/blob/main/config.json',
}
class _a ( snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Dict = 'open-llama'
def __init__( self : Optional[int] , UpperCAmelCase : Union[str, Any]=100000 , UpperCAmelCase : Tuple=4096 , UpperCAmelCase : int=11008 , UpperCAmelCase : Tuple=32 , UpperCAmelCase : Tuple=32 , UpperCAmelCase : Optional[Any]="silu" , UpperCAmelCase : int=2048 , UpperCAmelCase : Optional[int]=0.02 , UpperCAmelCase : Optional[Any]=1E-6 , UpperCAmelCase : List[Any]=True , UpperCAmelCase : Tuple=0 , UpperCAmelCase : Tuple=1 , UpperCAmelCase : Tuple=2 , UpperCAmelCase : Union[str, Any]=False , UpperCAmelCase : Optional[int]=True , UpperCAmelCase : Optional[int]=0.1 , UpperCAmelCase : str=0.1 , UpperCAmelCase : int=True , UpperCAmelCase : Optional[int]=True , UpperCAmelCase : List[str]=None , **UpperCAmelCase : str , ):
A_ = vocab_size
A_ = max_position_embeddings
A_ = hidden_size
A_ = intermediate_size
A_ = num_hidden_layers
A_ = num_attention_heads
A_ = hidden_act
A_ = initializer_range
A_ = rms_norm_eps
A_ = use_cache
A_ = kwargs.pop(
"use_memorry_efficient_attention" , UpperCAmelCase )
A_ = hidden_dropout_prob
A_ = attention_dropout_prob
A_ = use_stable_embedding
A_ = shared_input_output_embedding
A_ = rope_scaling
self._rope_scaling_validation()
super().__init__(
pad_token_id=UpperCAmelCase , bos_token_id=UpperCAmelCase , eos_token_id=UpperCAmelCase , tie_word_embeddings=UpperCAmelCase , **UpperCAmelCase , )
def __A ( self : Dict ):
if self.rope_scaling is None:
return
if not isinstance(self.rope_scaling , UpperCAmelCase ) or len(self.rope_scaling ) != 2:
raise ValueError(
"`rope_scaling` must be a dictionary with with two fields, `name` and `factor`, "
f'''got {self.rope_scaling}''' )
A_ = self.rope_scaling.get("type" , UpperCAmelCase )
A_ = self.rope_scaling.get("factor" , UpperCAmelCase )
if rope_scaling_type is None or rope_scaling_type not in ["linear", "dynamic"]:
raise ValueError(
f'''`rope_scaling`\'s name field must be one of [\'linear\', \'dynamic\'], got {rope_scaling_type}''' )
if rope_scaling_factor is None or not isinstance(UpperCAmelCase , UpperCAmelCase ) or rope_scaling_factor <= 1.0:
raise ValueError(f'''`rope_scaling`\'s factor field must be an float > 1, got {rope_scaling_factor}''' ) | 86 |
import argparse
import os
from transformers.utils import direct_transformers_import
# All paths are set with the intent you should run this script from the root of the repo with the command
# python utils/check_task_guides.py
__a :Optional[Any] = 'src/transformers'
__a :Tuple = 'docs/source/en/tasks'
def __snake_case ( __UpperCamelCase : List[Any] ,__UpperCamelCase : Dict ,__UpperCamelCase : int ):
"""simple docstring"""
with open(__UpperCamelCase ,"r" ,encoding="utf-8" ,newline="\n" ) as f:
A_ = f.readlines()
# Find the start prompt.
A_ = 0
while not lines[start_index].startswith(__UpperCamelCase ):
start_index += 1
start_index += 1
A_ = start_index
while not lines[end_index].startswith(__UpperCamelCase ):
end_index += 1
end_index -= 1
while len(lines[start_index] ) <= 1:
start_index += 1
while len(lines[end_index] ) <= 1:
end_index -= 1
end_index += 1
return "".join(lines[start_index:end_index] ), start_index, end_index, lines
# This is to make sure the transformers module imported is the one in the repo.
__a :List[str] = direct_transformers_import(TRANSFORMERS_PATH)
__a :Optional[Any] = {
'asr.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_CTC_MAPPING_NAMES,
'audio_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_AUDIO_CLASSIFICATION_MAPPING_NAMES,
'language_modeling.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_CAUSAL_LM_MAPPING_NAMES,
'image_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_IMAGE_CLASSIFICATION_MAPPING_NAMES,
'masked_language_modeling.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_MASKED_LM_MAPPING_NAMES,
'multiple_choice.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_MULTIPLE_CHOICE_MAPPING_NAMES,
'object_detection.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_OBJECT_DETECTION_MAPPING_NAMES,
'question_answering.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_QUESTION_ANSWERING_MAPPING_NAMES,
'semantic_segmentation.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEMANTIC_SEGMENTATION_MAPPING_NAMES,
'sequence_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING_NAMES,
'summarization.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES,
'token_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING_NAMES,
'translation.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES,
'video_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_VIDEO_CLASSIFICATION_MAPPING_NAMES,
'document_question_answering.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_DOCUMENT_QUESTION_ANSWERING_MAPPING_NAMES,
'monocular_depth_estimation.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_DEPTH_ESTIMATION_MAPPING_NAMES,
}
# This list contains model types used in some task guides that are not in `CONFIG_MAPPING_NAMES` (therefore not in any
# `MODEL_MAPPING_NAMES` or any `MODEL_FOR_XXX_MAPPING_NAMES`).
__a :Optional[Any] = {
'summarization.md': ('nllb',),
'translation.md': ('nllb',),
}
def __snake_case ( __UpperCamelCase : Tuple ):
"""simple docstring"""
A_ = TASK_GUIDE_TO_MODELS[task_guide]
A_ = SPECIAL_TASK_GUIDE_TO_MODEL_TYPES.get(__UpperCamelCase ,set() )
A_ = {
code: name
for code, name in transformers_module.MODEL_NAMES_MAPPING.items()
if (code in model_maping_names or code in special_model_types)
}
return ", ".join([f'''[{name}](../model_doc/{code})''' for code, name in model_names.items()] ) + "\n"
def __snake_case ( __UpperCamelCase : str ,__UpperCamelCase : List[str]=False ):
"""simple docstring"""
A_ , A_ , A_ , A_ = _find_text_in_file(
filename=os.path.join(__UpperCamelCase ,__UpperCamelCase ) ,start_prompt="<!--This tip is automatically generated by `make fix-copies`, do not fill manually!-->" ,end_prompt="<!--End of the generated tip-->" ,)
A_ = get_model_list_for_task(__UpperCamelCase )
if current_list != new_list:
if overwrite:
with open(os.path.join(__UpperCamelCase ,__UpperCamelCase ) ,"w" ,encoding="utf-8" ,newline="\n" ) as f:
f.writelines(lines[:start_index] + [new_list] + lines[end_index:] )
else:
raise ValueError(
f'''The list of models that can be used in the {task_guide} guide needs an update. Run `make fix-copies`'''
" to fix this." )
if __name__ == "__main__":
__a :int = argparse.ArgumentParser()
parser.add_argument('--fix_and_overwrite', action='store_true', help='Whether to fix inconsistencies.')
__a :Optional[Any] = parser.parse_args()
for task_guide in TASK_GUIDE_TO_MODELS.keys():
check_model_list_for_task(task_guide, args.fix_and_overwrite) | 86 | 1 |
import random
from typing import Any
def __snake_case ( __UpperCamelCase : list ):
"""simple docstring"""
for _ in range(len(__UpperCamelCase ) ):
A_ = random.randint(0 ,len(__UpperCamelCase ) - 1 )
A_ = random.randint(0 ,len(__UpperCamelCase ) - 1 )
A_ , A_ = data[b], data[a]
return data
if __name__ == "__main__":
__a :List[Any] = [0, 1, 2, 3, 4, 5, 6, 7]
__a :List[str] = ['python', 'says', 'hello', '!']
print('Fisher-Yates Shuffle:')
print('List', integers, strings)
print('FY Shuffle', fisher_yates_shuffle(integers), fisher_yates_shuffle(strings)) | 86 |
import argparse
import json
from pathlib import Path
import requests
import timm
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from transformers import DeiTImageProcessor, ViTConfig, ViTForImageClassification, ViTImageProcessor, ViTModel
from transformers.utils import logging
logging.set_verbosity_info()
__a :Dict = logging.get_logger(__name__)
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : Tuple=False ):
"""simple docstring"""
A_ = []
for i in range(config.num_hidden_layers ):
# encoder layers: output projection, 2 feedforward neural networks and 2 layernorms
rename_keys.append((f'''blocks.{i}.norm1.weight''', f'''vit.encoder.layer.{i}.layernorm_before.weight''') )
rename_keys.append((f'''blocks.{i}.norm1.bias''', f'''vit.encoder.layer.{i}.layernorm_before.bias''') )
rename_keys.append((f'''blocks.{i}.attn.proj.weight''', f'''vit.encoder.layer.{i}.attention.output.dense.weight''') )
rename_keys.append((f'''blocks.{i}.attn.proj.bias''', f'''vit.encoder.layer.{i}.attention.output.dense.bias''') )
rename_keys.append((f'''blocks.{i}.norm2.weight''', f'''vit.encoder.layer.{i}.layernorm_after.weight''') )
rename_keys.append((f'''blocks.{i}.norm2.bias''', f'''vit.encoder.layer.{i}.layernorm_after.bias''') )
rename_keys.append((f'''blocks.{i}.mlp.fc1.weight''', f'''vit.encoder.layer.{i}.intermediate.dense.weight''') )
rename_keys.append((f'''blocks.{i}.mlp.fc1.bias''', f'''vit.encoder.layer.{i}.intermediate.dense.bias''') )
rename_keys.append((f'''blocks.{i}.mlp.fc2.weight''', f'''vit.encoder.layer.{i}.output.dense.weight''') )
rename_keys.append((f'''blocks.{i}.mlp.fc2.bias''', f'''vit.encoder.layer.{i}.output.dense.bias''') )
# projection layer + position embeddings
rename_keys.extend(
[
("cls_token", "vit.embeddings.cls_token"),
("patch_embed.proj.weight", "vit.embeddings.patch_embeddings.projection.weight"),
("patch_embed.proj.bias", "vit.embeddings.patch_embeddings.projection.bias"),
("pos_embed", "vit.embeddings.position_embeddings"),
] )
if base_model:
# layernorm + pooler
rename_keys.extend(
[
("norm.weight", "layernorm.weight"),
("norm.bias", "layernorm.bias"),
("pre_logits.fc.weight", "pooler.dense.weight"),
("pre_logits.fc.bias", "pooler.dense.bias"),
] )
# if just the base model, we should remove "vit" from all keys that start with "vit"
A_ = [(pair[0], pair[1][4:]) if pair[1].startswith("vit" ) else pair for pair in rename_keys]
else:
# layernorm + classification head
rename_keys.extend(
[
("norm.weight", "vit.layernorm.weight"),
("norm.bias", "vit.layernorm.bias"),
("head.weight", "classifier.weight"),
("head.bias", "classifier.bias"),
] )
return rename_keys
def __snake_case ( __UpperCamelCase : Optional[int] ,__UpperCamelCase : Tuple ,__UpperCamelCase : Any=False ):
"""simple docstring"""
for i in range(config.num_hidden_layers ):
if base_model:
A_ = ""
else:
A_ = "vit."
# read in weights + bias of input projection layer (in timm, this is a single matrix + bias)
A_ = state_dict.pop(f'''blocks.{i}.attn.qkv.weight''' )
A_ = state_dict.pop(f'''blocks.{i}.attn.qkv.bias''' )
# next, add query, keys and values (in that order) to the state dict
A_ = in_proj_weight[
: config.hidden_size, :
]
A_ = in_proj_bias[: config.hidden_size]
A_ = in_proj_weight[
config.hidden_size : config.hidden_size * 2, :
]
A_ = in_proj_bias[
config.hidden_size : config.hidden_size * 2
]
A_ = in_proj_weight[
-config.hidden_size :, :
]
A_ = in_proj_bias[-config.hidden_size :]
def __snake_case ( __UpperCamelCase : List[Any] ):
"""simple docstring"""
A_ = ["head.weight", "head.bias"]
for k in ignore_keys:
state_dict.pop(__UpperCamelCase ,__UpperCamelCase )
def __snake_case ( __UpperCamelCase : Any ,__UpperCamelCase : Optional[int] ,__UpperCamelCase : List[str] ):
"""simple docstring"""
A_ = dct.pop(__UpperCamelCase )
A_ = val
def __snake_case ( ):
"""simple docstring"""
A_ = "http://images.cocodataset.org/val2017/000000039769.jpg"
A_ = Image.open(requests.get(__UpperCamelCase ,stream=__UpperCamelCase ).raw )
return im
@torch.no_grad()
def __snake_case ( __UpperCamelCase : Union[str, Any] ,__UpperCamelCase : Optional[int] ):
"""simple docstring"""
A_ = ViTConfig()
A_ = False
# dataset (ImageNet-21k only or also fine-tuned on ImageNet 2012), patch_size and image_size
if vit_name[-5:] == "in21k":
A_ = True
A_ = int(vit_name[-12:-10] )
A_ = int(vit_name[-9:-6] )
else:
A_ = 1000
A_ = "huggingface/label-files"
A_ = "imagenet-1k-id2label.json"
A_ = json.load(open(hf_hub_download(__UpperCamelCase ,__UpperCamelCase ,repo_type="dataset" ) ,"r" ) )
A_ = {int(__UpperCamelCase ): v for k, v in idalabel.items()}
A_ = idalabel
A_ = {v: k for k, v in idalabel.items()}
A_ = int(vit_name[-6:-4] )
A_ = int(vit_name[-3:] )
# size of the architecture
if "deit" in vit_name:
if vit_name[9:].startswith("tiny" ):
A_ = 192
A_ = 768
A_ = 12
A_ = 3
elif vit_name[9:].startswith("small" ):
A_ = 384
A_ = 1536
A_ = 12
A_ = 6
else:
pass
else:
if vit_name[4:].startswith("small" ):
A_ = 768
A_ = 2304
A_ = 8
A_ = 8
elif vit_name[4:].startswith("base" ):
pass
elif vit_name[4:].startswith("large" ):
A_ = 1024
A_ = 4096
A_ = 24
A_ = 16
elif vit_name[4:].startswith("huge" ):
A_ = 1280
A_ = 5120
A_ = 32
A_ = 16
# load original model from timm
A_ = timm.create_model(__UpperCamelCase ,pretrained=__UpperCamelCase )
timm_model.eval()
# load state_dict of original model, remove and rename some keys
A_ = timm_model.state_dict()
if base_model:
remove_classification_head_(__UpperCamelCase )
A_ = create_rename_keys(__UpperCamelCase ,__UpperCamelCase )
for src, dest in rename_keys:
rename_key(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
read_in_q_k_v(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
# load HuggingFace model
if vit_name[-5:] == "in21k":
A_ = ViTModel(__UpperCamelCase ).eval()
else:
A_ = ViTForImageClassification(__UpperCamelCase ).eval()
model.load_state_dict(__UpperCamelCase )
# Check outputs on an image, prepared by ViTImageProcessor/DeiTImageProcessor
if "deit" in vit_name:
A_ = DeiTImageProcessor(size=config.image_size )
else:
A_ = ViTImageProcessor(size=config.image_size )
A_ = image_processor(images=prepare_img() ,return_tensors="pt" )
A_ = encoding["pixel_values"]
A_ = model(__UpperCamelCase )
if base_model:
A_ = timm_model.forward_features(__UpperCamelCase )
assert timm_pooled_output.shape == outputs.pooler_output.shape
assert torch.allclose(__UpperCamelCase ,outputs.pooler_output ,atol=1E-3 )
else:
A_ = timm_model(__UpperCamelCase )
assert timm_logits.shape == outputs.logits.shape
assert torch.allclose(__UpperCamelCase ,outputs.logits ,atol=1E-3 )
Path(__UpperCamelCase ).mkdir(exist_ok=__UpperCamelCase )
print(f'''Saving model {vit_name} to {pytorch_dump_folder_path}''' )
model.save_pretrained(__UpperCamelCase )
print(f'''Saving image processor to {pytorch_dump_folder_path}''' )
image_processor.save_pretrained(__UpperCamelCase )
if __name__ == "__main__":
__a :str = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
'--vit_name',
default='vit_base_patch16_224',
type=str,
help='Name of the ViT timm model you\'d like to convert.',
)
parser.add_argument(
'--pytorch_dump_folder_path', default=None, type=str, help='Path to the output PyTorch model directory.'
)
__a :Optional[int] = parser.parse_args()
convert_vit_checkpoint(args.vit_name, args.pytorch_dump_folder_path) | 86 | 1 |
__a :Tuple = '0.21.0'
from .accelerator import Accelerator
from .big_modeling import (
cpu_offload,
cpu_offload_with_hook,
disk_offload,
dispatch_model,
init_empty_weights,
init_on_device,
load_checkpoint_and_dispatch,
)
from .data_loader import skip_first_batches
from .launchers import debug_launcher, notebook_launcher
from .state import PartialState
from .utils import (
DeepSpeedPlugin,
DistributedDataParallelKwargs,
DistributedType,
FullyShardedDataParallelPlugin,
GradScalerKwargs,
InitProcessGroupKwargs,
find_executable_batch_size,
infer_auto_device_map,
is_rich_available,
load_checkpoint_in_model,
synchronize_rng_states,
)
if is_rich_available():
from .utils import rich | 86 |
def __snake_case ( __UpperCamelCase : int = 50 ):
"""simple docstring"""
A_ = [1] * (length + 1)
for row_length in range(length + 1 ):
for tile_length in range(2 ,5 ):
for tile_start in range(row_length - tile_length + 1 ):
ways_number[row_length] += ways_number[
row_length - tile_start - tile_length
]
return ways_number[length]
if __name__ == "__main__":
print(F"{solution() = }") | 86 | 1 |
def __snake_case ( __UpperCamelCase : Union[str, Any] ,__UpperCamelCase : Dict ):
"""simple docstring"""
A_ = [0 for i in range(r + 1 )]
# nc0 = 1
A_ = 1
for i in range(1 ,n + 1 ):
# to compute current row from previous row.
A_ = min(__UpperCamelCase ,__UpperCamelCase )
while j > 0:
c[j] += c[j - 1]
j -= 1
return c[r]
print(binomial_coefficient(n=10, r=5)) | 86 |
from typing import Any, Dict, List, Union
from ..utils import add_end_docstrings, is_torch_available, is_vision_available, logging, requires_backends
from .base import PIPELINE_INIT_ARGS, ChunkPipeline
if is_vision_available():
from PIL import Image
from ..image_utils import load_image
if is_torch_available():
import torch
from transformers.modeling_outputs import BaseModelOutput
from ..models.auto.modeling_auto import MODEL_FOR_ZERO_SHOT_OBJECT_DETECTION_MAPPING
__a :List[str] = logging.get_logger(__name__)
@add_end_docstrings(snake_case_ )
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Any , **UpperCAmelCase : List[str] ):
super().__init__(**UpperCAmelCase )
if self.framework == "tf":
raise ValueError(f'''The {self.__class__} is only available in PyTorch.''' )
requires_backends(self , "vision" )
self.check_model_type(UpperCAmelCase )
def __call__( self : Optional[int] , UpperCAmelCase : Union[str, "Image.Image", List[Dict[str, Any]]] , UpperCAmelCase : Union[str, List[str]] = None , **UpperCAmelCase : List[Any] , ):
if "text_queries" in kwargs:
A_ = kwargs.pop("text_queries" )
if isinstance(UpperCAmelCase , (str, Image.Image) ):
A_ = {"image": image, "candidate_labels": candidate_labels}
else:
A_ = image
A_ = super().__call__(UpperCAmelCase , **UpperCAmelCase )
return results
def __A ( self : int , **UpperCAmelCase : Tuple ):
A_ = {}
if "threshold" in kwargs:
A_ = kwargs["threshold"]
if "top_k" in kwargs:
A_ = kwargs["top_k"]
return {}, {}, postprocess_params
def __A ( self : List[str] , UpperCAmelCase : Dict ):
A_ = load_image(inputs["image"] )
A_ = inputs["candidate_labels"]
if isinstance(UpperCAmelCase , UpperCAmelCase ):
A_ = candidate_labels.split("," )
A_ = torch.tensor([[image.height, image.width]] , dtype=torch.intaa )
for i, candidate_label in enumerate(UpperCAmelCase ):
A_ = self.tokenizer(UpperCAmelCase , return_tensors=self.framework )
A_ = self.image_processor(UpperCAmelCase , return_tensors=self.framework )
yield {
"is_last": i == len(UpperCAmelCase ) - 1,
"target_size": target_size,
"candidate_label": candidate_label,
**text_inputs,
**image_features,
}
def __A ( self : str , UpperCAmelCase : int ):
A_ = model_inputs.pop("target_size" )
A_ = model_inputs.pop("candidate_label" )
A_ = model_inputs.pop("is_last" )
A_ = self.model(**UpperCAmelCase )
A_ = {"target_size": target_size, "candidate_label": candidate_label, "is_last": is_last, **outputs}
return model_outputs
def __A ( self : Dict , UpperCAmelCase : Any , UpperCAmelCase : Optional[int]=0.1 , UpperCAmelCase : Optional[int]=None ):
A_ = []
for model_output in model_outputs:
A_ = model_output["candidate_label"]
A_ = BaseModelOutput(UpperCAmelCase )
A_ = self.image_processor.post_process_object_detection(
outputs=UpperCAmelCase , threshold=UpperCAmelCase , target_sizes=model_output["target_size"] )[0]
for index in outputs["scores"].nonzero():
A_ = outputs["scores"][index].item()
A_ = self._get_bounding_box(outputs["boxes"][index][0] )
A_ = {"score": score, "label": label, "box": box}
results.append(UpperCAmelCase )
A_ = sorted(UpperCAmelCase , key=lambda UpperCAmelCase : x["score"] , reverse=UpperCAmelCase )
if top_k:
A_ = results[:top_k]
return results
def __A ( self : List[str] , UpperCAmelCase : "torch.Tensor" ):
if self.framework != "pt":
raise ValueError("The ZeroShotObjectDetectionPipeline is only available in PyTorch." )
A_ , A_ , A_ , A_ = box.int().tolist()
A_ = {
"xmin": xmin,
"ymin": ymin,
"xmax": xmax,
"ymax": ymax,
}
return bbox | 86 | 1 |
import json
import os
import unittest
from transformers import BatchEncoding, LEDTokenizer, LEDTokenizerFast
from transformers.models.led.tokenization_led import VOCAB_FILES_NAMES
from transformers.testing_utils import require_tokenizers, require_torch
from transformers.utils import cached_property
from ...test_tokenization_common import TokenizerTesterMixin
@require_tokenizers
class _a ( snake_case_ , unittest.TestCase ):
"""simple docstring"""
_lowerCamelCase : int = LEDTokenizer
_lowerCamelCase : str = LEDTokenizerFast
_lowerCamelCase : List[Any] = True
def __A ( self : Optional[int] ):
super().setUp()
A_ = [
"l",
"o",
"w",
"e",
"r",
"s",
"t",
"i",
"d",
"n",
"\u0120",
"\u0120l",
"\u0120n",
"\u0120lo",
"\u0120low",
"er",
"\u0120lowest",
"\u0120newer",
"\u0120wider",
"<unk>",
]
A_ = dict(zip(UpperCAmelCase , range(len(UpperCAmelCase ) ) ) )
A_ = ["#version: 0.2", "\u0120 l", "\u0120l o", "\u0120lo w", "e r", ""]
A_ = {"unk_token": "<unk>"}
A_ = os.path.join(self.tmpdirname , VOCAB_FILES_NAMES["vocab_file"] )
A_ = os.path.join(self.tmpdirname , VOCAB_FILES_NAMES["merges_file"] )
with open(self.vocab_file , "w" , encoding="utf-8" ) as fp:
fp.write(json.dumps(UpperCAmelCase ) + "\n" )
with open(self.merges_file , "w" , encoding="utf-8" ) as fp:
fp.write("\n".join(UpperCAmelCase ) )
def __A ( self : Optional[Any] , **UpperCAmelCase : Optional[int] ):
kwargs.update(self.special_tokens_map )
return self.tokenizer_class.from_pretrained(self.tmpdirname , **UpperCAmelCase )
def __A ( self : Tuple , **UpperCAmelCase : int ):
kwargs.update(self.special_tokens_map )
return self.rust_tokenizer_class.from_pretrained(self.tmpdirname , **UpperCAmelCase )
def __A ( self : Any , UpperCAmelCase : Dict ):
return "lower newer", "lower newer"
@cached_property
def __A ( self : Any ):
return LEDTokenizer.from_pretrained("allenai/led-base-16384" )
@cached_property
def __A ( self : Union[str, Any] ):
return LEDTokenizerFast.from_pretrained("allenai/led-base-16384" )
@require_torch
def __A ( self : str ):
A_ = ["A long paragraph for summarization.", "Another paragraph for summarization."]
A_ = [0, 250, 251, 17818, 13, 39186, 1938, 4, 2]
for tokenizer in [self.default_tokenizer, self.default_tokenizer_fast]:
A_ = tokenizer(UpperCAmelCase , max_length=len(UpperCAmelCase ) , padding=UpperCAmelCase , return_tensors="pt" )
self.assertIsInstance(UpperCAmelCase , UpperCAmelCase )
self.assertEqual((2, 9) , batch.input_ids.shape )
self.assertEqual((2, 9) , batch.attention_mask.shape )
A_ = batch.input_ids.tolist()[0]
self.assertListEqual(UpperCAmelCase , UpperCAmelCase )
@require_torch
def __A ( self : Optional[int] ):
A_ = ["A long paragraph for summarization.", "Another paragraph for summarization."]
for tokenizer in [self.default_tokenizer, self.default_tokenizer_fast]:
A_ = tokenizer(UpperCAmelCase , padding=UpperCAmelCase , return_tensors="pt" )
self.assertIn("input_ids" , UpperCAmelCase )
self.assertIn("attention_mask" , UpperCAmelCase )
self.assertNotIn("labels" , UpperCAmelCase )
self.assertNotIn("decoder_attention_mask" , UpperCAmelCase )
@require_torch
def __A ( self : Tuple ):
A_ = [
"Summary of the text.",
"Another summary.",
]
for tokenizer in [self.default_tokenizer, self.default_tokenizer_fast]:
A_ = tokenizer(text_target=UpperCAmelCase , max_length=32 , padding="max_length" , return_tensors="pt" )
self.assertEqual(32 , targets["input_ids"].shape[1] )
@require_torch
def __A ( self : List[Any] ):
for tokenizer in [self.default_tokenizer, self.default_tokenizer_fast]:
A_ = tokenizer(
["I am a small frog" * 1024, "I am a small frog"] , padding=UpperCAmelCase , truncation=UpperCAmelCase , return_tensors="pt" )
self.assertIsInstance(UpperCAmelCase , UpperCAmelCase )
self.assertEqual(batch.input_ids.shape , (2, 5122) )
@require_torch
def __A ( self : Any ):
A_ = ["A long paragraph for summarization."]
A_ = [
"Summary of the text.",
]
for tokenizer in [self.default_tokenizer, self.default_tokenizer_fast]:
A_ = tokenizer(UpperCAmelCase , return_tensors="pt" )
A_ = tokenizer(text_target=UpperCAmelCase , return_tensors="pt" )
A_ = inputs["input_ids"]
A_ = targets["input_ids"]
self.assertTrue((input_ids[:, 0] == tokenizer.bos_token_id).all().item() )
self.assertTrue((labels[:, 0] == tokenizer.bos_token_id).all().item() )
self.assertTrue((input_ids[:, -1] == tokenizer.eos_token_id).all().item() )
self.assertTrue((labels[:, -1] == tokenizer.eos_token_id).all().item() )
@require_torch
def __A ( self : List[str] ):
for tokenizer in [self.default_tokenizer, self.default_tokenizer_fast]:
A_ = ["Summary of the text.", "Another summary."]
A_ = [[0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, -1, -1]]
A_ = tokenizer(UpperCAmelCase , padding=UpperCAmelCase )
A_ = [[0] * len(UpperCAmelCase ) for x in encoded_output["input_ids"]]
A_ = tokenizer.pad(UpperCAmelCase )
self.assertSequenceEqual(outputs["global_attention_mask"] , UpperCAmelCase )
def __A ( self : str ):
pass
def __A ( self : Tuple ):
for tokenizer, pretrained_name, kwargs in self.tokenizers_list:
with self.subTest(f'''{tokenizer.__class__.__name__} ({pretrained_name})''' ):
A_ = self.rust_tokenizer_class.from_pretrained(UpperCAmelCase , **UpperCAmelCase )
A_ = self.tokenizer_class.from_pretrained(UpperCAmelCase , **UpperCAmelCase )
A_ = "A, <mask> AllenNLP sentence."
A_ = tokenizer_r.encode_plus(UpperCAmelCase , add_special_tokens=UpperCAmelCase , return_token_type_ids=UpperCAmelCase )
A_ = tokenizer_p.encode_plus(UpperCAmelCase , add_special_tokens=UpperCAmelCase , return_token_type_ids=UpperCAmelCase )
self.assertEqual(sum(tokens_r["token_type_ids"] ) , sum(tokens_p["token_type_ids"] ) )
self.assertEqual(
sum(tokens_r["attention_mask"] ) / len(tokens_r["attention_mask"] ) , sum(tokens_p["attention_mask"] ) / len(tokens_p["attention_mask"] ) , )
A_ = tokenizer_r.convert_ids_to_tokens(tokens_r["input_ids"] )
A_ = tokenizer_p.convert_ids_to_tokens(tokens_p["input_ids"] )
self.assertSequenceEqual(tokens_p["input_ids"] , [0, 250, 6, 50264, 3823, 487, 21992, 3645, 4, 2] )
self.assertSequenceEqual(tokens_r["input_ids"] , [0, 250, 6, 50264, 3823, 487, 21992, 3645, 4, 2] )
self.assertSequenceEqual(
UpperCAmelCase , ["<s>", "A", ",", "<mask>", "ĠAllen", "N", "LP", "Ġsentence", ".", "</s>"] )
self.assertSequenceEqual(
UpperCAmelCase , ["<s>", "A", ",", "<mask>", "ĠAllen", "N", "LP", "Ġsentence", ".", "</s>"] ) | 86 |
import argparse
import torch
# Step 1. clone https://github.com/microsoft/unilm
# Step 2. git checkout to https://github.com/microsoft/unilm/commit/b94ec76c36f02fb2b0bf0dcb0b8554a2185173cd
# Step 3. cd unilm
# Step 4. ln -s $(realpath wavlm/modules.py) ./ # create simlink
# import classes
from unilm.wavlm.WavLM import WavLM as WavLMOrig
from unilm.wavlm.WavLM import WavLMConfig as WavLMConfigOrig
from transformers import WavLMConfig, WavLMModel, logging
logging.set_verbosity_info()
__a :Any = logging.get_logger(__name__)
__a :int = {
'post_extract_proj': 'feature_projection.projection',
'encoder.pos_conv.0': 'encoder.pos_conv_embed.conv',
'self_attn.k_proj': 'encoder.layers.*.attention.k_proj',
'self_attn.v_proj': 'encoder.layers.*.attention.v_proj',
'self_attn.q_proj': 'encoder.layers.*.attention.q_proj',
'self_attn.out_proj': 'encoder.layers.*.attention.out_proj',
'self_attn.grep_linear': 'encoder.layers.*.attention.gru_rel_pos_linear',
'self_attn.relative_attention_bias': 'encoder.layers.*.attention.rel_attn_embed',
'self_attn.grep_a': 'encoder.layers.*.attention.gru_rel_pos_const',
'self_attn_layer_norm': 'encoder.layers.*.layer_norm',
'fc1': 'encoder.layers.*.feed_forward.intermediate_dense',
'fc2': 'encoder.layers.*.feed_forward.output_dense',
'final_layer_norm': 'encoder.layers.*.final_layer_norm',
'encoder.layer_norm': 'encoder.layer_norm',
'w2v_model.layer_norm': 'feature_projection.layer_norm',
'quantizer.weight_proj': 'quantizer.weight_proj',
'quantizer.vars': 'quantizer.codevectors',
'project_q': 'project_q',
'final_proj': 'project_hid',
'w2v_encoder.proj': 'ctc_proj',
'mask_emb': 'masked_spec_embed',
}
__a :Tuple = [
'ctc_proj',
'quantizer.weight_proj',
'quantizer.codevectors',
'project_q',
'project_hid',
]
def __snake_case ( __UpperCamelCase : List[str] ,__UpperCamelCase : List[Any] ,__UpperCamelCase : Dict ,__UpperCamelCase : Any ,__UpperCamelCase : Optional[int] ):
"""simple docstring"""
for attribute in key.split("." ):
A_ = getattr(__UpperCamelCase ,__UpperCamelCase )
if weight_type is not None:
A_ = getattr(__UpperCamelCase ,__UpperCamelCase ).shape
else:
A_ = hf_pointer.shape
assert hf_shape == value.shape, (
f'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
f''' {value.shape} for {full_name}'''
)
if weight_type == "weight":
A_ = value
elif weight_type == "weight_g":
A_ = value
elif weight_type == "weight_v":
A_ = value
elif weight_type == "bias":
A_ = value
else:
A_ = value
logger.info(f'''{key + "." + weight_type if weight_type is not None else ""} was initialized from {full_name}.''' )
def __snake_case ( __UpperCamelCase : Optional[int] ,__UpperCamelCase : Optional[Any] ):
"""simple docstring"""
A_ = []
A_ = fairseq_model.state_dict()
A_ = hf_model.feature_extractor
for name, value in fairseq_dict.items():
A_ = False
if "conv_layers" in name:
load_conv_layer(
__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,hf_model.config.feat_extract_norm == "group" ,)
A_ = True
else:
for key, mapped_key in MAPPING.items():
if key in name or key.split("w2v_model." )[-1] == name.split("." )[0]:
A_ = True
if "*" in mapped_key:
A_ = name.split(__UpperCamelCase )[0].split("." )[-2]
A_ = mapped_key.replace("*" ,__UpperCamelCase )
if "weight_g" in name:
A_ = "weight_g"
elif "weight_v" in name:
A_ = "weight_v"
elif "bias" in name and "relative_attention_bias" not in name:
A_ = "bias"
elif "weight" in name:
# TODO: don't match quantizer.weight_proj
A_ = "weight"
else:
A_ = None
set_recursively(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
continue
if not is_used:
unused_weights.append(__UpperCamelCase )
logger.warning(f'''Unused weights: {unused_weights}''' )
def __snake_case ( __UpperCamelCase : Optional[int] ,__UpperCamelCase : Dict ,__UpperCamelCase : List[str] ,__UpperCamelCase : Union[str, Any] ,__UpperCamelCase : Optional[int] ):
"""simple docstring"""
A_ = full_name.split("conv_layers." )[-1]
A_ = name.split("." )
A_ = int(items[0] )
A_ = int(items[1] )
if type_id == 0:
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.bias.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape, (
f'''{full_name} has size {value.shape}, but {feature_extractor[layer_id].layer_norm.bias.data.shape} was'''
" found."
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor[layer_id].layer_norm.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(__UpperCamelCase )
@torch.no_grad()
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : str ,__UpperCamelCase : int=None ):
"""simple docstring"""
A_ = torch.load(__UpperCamelCase )
A_ = WavLMConfigOrig(checkpoint["cfg"] )
A_ = WavLMOrig(__UpperCamelCase )
model.load_state_dict(checkpoint["model"] )
model.eval()
if config_path is not None:
A_ = WavLMConfig.from_pretrained(__UpperCamelCase )
else:
A_ = WavLMConfig()
A_ = WavLMModel(__UpperCamelCase )
recursively_load_weights(__UpperCamelCase ,__UpperCamelCase )
hf_wavlm.save_pretrained(__UpperCamelCase )
if __name__ == "__main__":
__a :List[Any] = argparse.ArgumentParser()
parser.add_argument('--pytorch_dump_folder_path', default=None, type=str, help='Path to the output PyTorch model.')
parser.add_argument('--checkpoint_path', default=None, type=str, help='Path to fairseq checkpoint')
parser.add_argument('--config_path', default=None, type=str, help='Path to hf config.json of model to convert')
__a :Optional[int] = parser.parse_args()
convert_wavlm_checkpoint(args.checkpoint_path, args.pytorch_dump_folder_path, args.config_path) | 86 | 1 |
from __future__ import annotations
__a :List[Any] = list[list[int]]
# assigning initial values to the grid
__a :Matrix = [
[3, 0, 6, 5, 0, 8, 4, 0, 0],
[5, 2, 0, 0, 0, 0, 0, 0, 0],
[0, 8, 7, 0, 0, 0, 0, 3, 1],
[0, 0, 3, 0, 1, 0, 0, 8, 0],
[9, 0, 0, 8, 6, 3, 0, 0, 5],
[0, 5, 0, 0, 9, 0, 6, 0, 0],
[1, 3, 0, 0, 0, 0, 2, 5, 0],
[0, 0, 0, 0, 0, 0, 0, 7, 4],
[0, 0, 5, 2, 0, 6, 3, 0, 0],
]
# a grid with no solution
__a :Matrix = [
[5, 0, 6, 5, 0, 8, 4, 0, 3],
[5, 2, 0, 0, 0, 0, 0, 0, 2],
[1, 8, 7, 0, 0, 0, 0, 3, 1],
[0, 0, 3, 0, 1, 0, 0, 8, 0],
[9, 0, 0, 8, 6, 3, 0, 0, 5],
[0, 5, 0, 0, 9, 0, 6, 0, 0],
[1, 3, 0, 0, 0, 0, 2, 5, 0],
[0, 0, 0, 0, 0, 0, 0, 7, 4],
[0, 0, 5, 2, 0, 6, 3, 0, 0],
]
def __snake_case ( __UpperCamelCase : Matrix ,__UpperCamelCase : int ,__UpperCamelCase : int ,__UpperCamelCase : int ):
"""simple docstring"""
for i in range(9 ):
if grid[row][i] == n or grid[i][column] == n:
return False
for i in range(3 ):
for j in range(3 ):
if grid[(row - row % 3) + i][(column - column % 3) + j] == n:
return False
return True
def __snake_case ( __UpperCamelCase : Matrix ):
"""simple docstring"""
for i in range(9 ):
for j in range(9 ):
if grid[i][j] == 0:
return i, j
return None
def __snake_case ( __UpperCamelCase : Matrix ):
"""simple docstring"""
if location := find_empty_location(__UpperCamelCase ):
A_ , A_ = location
else:
# If the location is ``None``, then the grid is solved.
return grid
for digit in range(1 ,10 ):
if is_safe(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ):
A_ = digit
if sudoku(__UpperCamelCase ) is not None:
return grid
A_ = 0
return None
def __snake_case ( __UpperCamelCase : Matrix ):
"""simple docstring"""
for row in grid:
for cell in row:
print(__UpperCamelCase ,end=" " )
print()
if __name__ == "__main__":
# make a copy of grid so that you can compare with the unmodified grid
for example_grid in (initial_grid, no_solution):
print('\nExample grid:\n' + '=' * 20)
print_solution(example_grid)
print('\nExample grid solution:')
__a :int = sudoku(example_grid)
if solution is not None:
print_solution(solution)
else:
print('Cannot find a solution.') | 86 |
def __snake_case ( __UpperCamelCase : list ,__UpperCamelCase : int = 0 ):
"""simple docstring"""
A_ = length or len(__UpperCamelCase )
A_ = False
for i in range(length - 1 ):
if list_data[i] > list_data[i + 1]:
A_ , A_ = list_data[i + 1], list_data[i]
A_ = True
return list_data if not swapped else bubble_sort(__UpperCamelCase ,length - 1 )
if __name__ == "__main__":
import doctest
doctest.testmod() | 86 | 1 |
def __snake_case ( __UpperCamelCase : list[list[int]] ,__UpperCamelCase : int ,__UpperCamelCase : int ,__UpperCamelCase : set ):
"""simple docstring"""
A_ , A_ = len(__UpperCamelCase ), len(grid[0] )
if (
min(__UpperCamelCase ,__UpperCamelCase ) < 0
or row == row_length
or col == col_length
or (row, col) in visit
or grid[row][col] == 1
):
return 0
if row == row_length - 1 and col == col_length - 1:
return 1
visit.add((row, col) )
A_ = 0
count += depth_first_search(__UpperCamelCase ,row + 1 ,__UpperCamelCase ,__UpperCamelCase )
count += depth_first_search(__UpperCamelCase ,row - 1 ,__UpperCamelCase ,__UpperCamelCase )
count += depth_first_search(__UpperCamelCase ,__UpperCamelCase ,col + 1 ,__UpperCamelCase )
count += depth_first_search(__UpperCamelCase ,__UpperCamelCase ,col - 1 ,__UpperCamelCase )
visit.remove((row, col) )
return count
if __name__ == "__main__":
import doctest
doctest.testmod() | 86 |
import pickle
import unittest
import torch
from accelerate import Accelerator
from accelerate.state import AcceleratorState
from accelerate.test_utils import require_cpu
@require_cpu
class _a ( unittest.TestCase ):
"""simple docstring"""
def __A ( self : List[str] ):
A_ = torch.nn.Linear(10 , 10 )
A_ = torch.optim.SGD(model.parameters() , 0.1 )
A_ = Accelerator()
A_ = accelerator.prepare(UpperCAmelCase )
try:
pickle.loads(pickle.dumps(UpperCAmelCase ) )
except Exception as e:
self.fail(f'''Accelerated optimizer pickling failed with {e}''' )
AcceleratorState._reset_state() | 86 | 1 |
import math
def __snake_case ( __UpperCamelCase : int ):
"""simple docstring"""
A_ = 0
A_ = 0
while num > 0:
A_ = num % 8
A_ = octal + (remainder * math.floor(math.pow(10 ,__UpperCamelCase ) ))
counter += 1
A_ = math.floor(num / 8 ) # basically /= 8 without remainder if any
# This formatting removes trailing '.0' from `octal`.
return f'''0o{int(__UpperCamelCase )}'''
def __snake_case ( ):
"""simple docstring"""
print("\n2 in octal is:" )
print(decimal_to_octal(2 ) ) # = 2
print("\n8 in octal is:" )
print(decimal_to_octal(8 ) ) # = 10
print("\n65 in octal is:" )
print(decimal_to_octal(65 ) ) # = 101
print("\n216 in octal is:" )
print(decimal_to_octal(216 ) ) # = 330
print("\n512 in octal is:" )
print(decimal_to_octal(512 ) ) # = 1000
print("\n" )
if __name__ == "__main__":
main() | 86 |
import argparse
import json
import os
import fairseq
import torch
from torch import nn
from transformers import (
SpeechaTextaConfig,
SpeechaTextaForCausalLM,
SpeechaTextaTokenizer,
SpeechEncoderDecoderConfig,
SpeechEncoderDecoderModel,
WavaVecaConfig,
WavaVecaFeatureExtractor,
WavaVecaModel,
logging,
)
logging.set_verbosity_info()
__a :List[str] = logging.get_logger(__name__)
__a :Optional[int] = {
'post_extract_proj': 'feature_projection.projection',
'encoder.pos_conv.0': 'encoder.pos_conv_embed.conv',
'self_attn.k_proj': 'encoder.layers.*.attention.k_proj',
'self_attn.v_proj': 'encoder.layers.*.attention.v_proj',
'self_attn.q_proj': 'encoder.layers.*.attention.q_proj',
'self_attn.out_proj': 'encoder.layers.*.attention.out_proj',
'self_attn_layer_norm': 'encoder.layers.*.layer_norm',
'fc1': 'encoder.layers.*.feed_forward.intermediate_dense',
'fc2': 'encoder.layers.*.feed_forward.output_dense',
'final_layer_norm': 'encoder.layers.*.final_layer_norm',
'encoder.layer_norm': 'encoder.layer_norm',
'w2v_model.layer_norm': 'feature_projection.layer_norm',
'quantizer.weight_proj': 'quantizer.weight_proj',
'quantizer.vars': 'quantizer.codevectors',
'project_q': 'project_q',
'final_proj': 'project_hid',
'w2v_encoder.proj': 'lm_head',
'mask_emb': 'masked_spec_embed',
}
__a :Any = [
'lm_head',
'quantizer.weight_proj',
'quantizer.codevectors',
'project_q',
'project_hid',
]
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : Optional[Any] ,__UpperCamelCase : Optional[int] ,__UpperCamelCase : List[str] ,__UpperCamelCase : Union[str, Any] ):
"""simple docstring"""
for attribute in key.split("." ):
A_ = getattr(__UpperCamelCase ,__UpperCamelCase )
if weight_type is not None:
A_ = getattr(__UpperCamelCase ,__UpperCamelCase ).shape
else:
A_ = hf_pointer.shape
assert hf_shape == value.shape, (
f'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
f''' {value.shape} for {full_name}'''
)
if weight_type == "weight":
A_ = value
elif weight_type == "weight_g":
A_ = value
elif weight_type == "weight_v":
A_ = value
elif weight_type == "bias":
A_ = value
else:
A_ = value
logger.info(f'''{key + "." + weight_type if weight_type is not None else ""} was initialized from {full_name}.''' )
def __snake_case ( __UpperCamelCase : Union[str, Any] ,__UpperCamelCase : Dict ):
"""simple docstring"""
A_ = []
A_ = fairseq_model.state_dict()
A_ = hf_model.feature_extractor
# if encoder has different dim to decoder -> use proj_weight
A_ = None
for name, value in fairseq_dict.items():
A_ = False
if "conv_layers" in name:
load_conv_layer(
__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,hf_model.config.feat_extract_norm == "group" ,)
A_ = True
elif name.split("." )[0] == "proj":
A_ = fairseq_model.proj
A_ = True
else:
for key, mapped_key in MAPPING.items():
if key in name or key.split("w2v_model." )[-1] == name.split("." )[0]:
A_ = True
if "*" in mapped_key:
A_ = name.split(__UpperCamelCase )[0].split("." )[-2]
A_ = mapped_key.replace("*" ,__UpperCamelCase )
if "weight_g" in name:
A_ = "weight_g"
elif "weight_v" in name:
A_ = "weight_v"
elif "bias" in name:
A_ = "bias"
elif "weight" in name:
A_ = "weight"
else:
A_ = None
set_recursively(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
continue
if not is_used:
unused_weights.append(__UpperCamelCase )
logger.warning(f'''Unused weights: {unused_weights}''' )
return proj_weight
def __snake_case ( __UpperCamelCase : str ,__UpperCamelCase : Optional[Any] ,__UpperCamelCase : int ,__UpperCamelCase : int ,__UpperCamelCase : Any ):
"""simple docstring"""
A_ = full_name.split("conv_layers." )[-1]
A_ = name.split("." )
A_ = int(items[0] )
A_ = int(items[1] )
if type_id == 0:
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.bias.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape, (
f'''{full_name} has size {value.shape}, but {feature_extractor[layer_id].layer_norm.bias.data.shape} was'''
" found."
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor[layer_id].layer_norm.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(__UpperCamelCase )
def __snake_case ( __UpperCamelCase : Optional[Any] ):
"""simple docstring"""
A_ , A_ = emb.weight.shape
A_ = nn.Linear(__UpperCamelCase ,__UpperCamelCase ,bias=__UpperCamelCase )
A_ = emb.weight.data
return lin_layer
def __snake_case ( __UpperCamelCase : Tuple ):
"""simple docstring"""
with open(__UpperCamelCase ,"r" ,encoding="utf-8" ) as f:
A_ = f.readlines()
A_ = [line.split(" " )[0] for line in lines]
A_ = len(__UpperCamelCase )
A_ = {
"<s>": 0,
"<pad>": 1,
"</s>": 2,
"<unk>": 3,
}
vocab_dict.update(dict(zip(__UpperCamelCase ,range(4 ,num_words + 4 ) ) ) )
return vocab_dict
@torch.no_grad()
def __snake_case ( __UpperCamelCase : List[str] ,__UpperCamelCase : Any ,__UpperCamelCase : List[Any] ,__UpperCamelCase : Union[str, Any] ,__UpperCamelCase : List[str] ,__UpperCamelCase : Optional[Any] ,__UpperCamelCase : Dict ,):
"""simple docstring"""
A_ = WavaVecaConfig.from_pretrained(__UpperCamelCase )
A_ = SpeechaTextaConfig.from_pretrained(
__UpperCamelCase ,vocab_size=__UpperCamelCase ,decoder_layers=__UpperCamelCase ,do_stable_layer_norm=__UpperCamelCase )
A_ = WavaVecaFeatureExtractor(
feature_size=1 ,sampling_rate=1_6000 ,padding_value=0 ,do_normalize=__UpperCamelCase ,return_attention_mask=__UpperCamelCase ,)
A_ , A_ , A_ = fairseq.checkpoint_utils.load_model_ensemble_and_task(
[checkpoint_path] ,arg_overrides={"data": "/".join(dict_path.split("/" )[:-1] )} )
A_ = model[0].eval()
# set weights for wav2vec2 encoder
A_ = WavaVecaModel(__UpperCamelCase )
A_ = recursively_load_weights_wavaveca(model.encoder ,__UpperCamelCase )
A_ = SpeechaTextaForCausalLM(__UpperCamelCase )
A_ , A_ = hf_decoder.model.decoder.load_state_dict(model.decoder.state_dict() ,strict=__UpperCamelCase )
# set output linear layer
unexpected_keys.remove("embed_out" )
A_ = nn.Parameter(model.decoder.embed_out.detach() )
# layer norm is init to identity matrix so leaving it is fine
logger.warning(f'''The following keys are missing when loading the decoder weights: {missing_keys}''' )
logger.warning(f'''The following keys are unexpected when loading the decoder weights: {unexpected_keys}''' )
A_ = SpeechEncoderDecoderModel(encoder=__UpperCamelCase ,decoder=__UpperCamelCase )
A_ = False
# add projection layer
A_ = nn.Parameter(projection_layer.weight )
A_ = nn.Parameter(projection_layer.bias )
A_ = create_vocab_dict(__UpperCamelCase )
with open(os.path.join(__UpperCamelCase ,"vocab.json" ) ,"w" ) as fp:
json.dump(__UpperCamelCase ,__UpperCamelCase )
A_ = SpeechaTextaTokenizer(os.path.join(__UpperCamelCase ,"vocab.json" ) )
tokenizer.save_pretrained(__UpperCamelCase )
A_ = hf_wavavec.config.to_dict()
A_ = tokenizer.pad_token_id
A_ = tokenizer.bos_token_id
A_ = tokenizer.eos_token_id
A_ = "speech_to_text_2"
A_ = "wav2vec2"
A_ = SpeechEncoderDecoderConfig.from_dict(__UpperCamelCase )
hf_wavavec.save_pretrained(__UpperCamelCase )
feature_extractor.save_pretrained(__UpperCamelCase )
if __name__ == "__main__":
__a :int = argparse.ArgumentParser()
parser.add_argument('--pytorch_dump_folder_path', default=None, type=str, help='Path to the output PyTorch model.')
parser.add_argument('--checkpoint_path', default=None, type=str, help='Path to fairseq checkpoint')
parser.add_argument('--dict_path', default=None, type=str, help='Path to dict of fine-tuned model')
parser.add_argument(
'--encoder_config_path',
default='facebook/wav2vec2-large-lv60',
type=str,
help='Path to hf encoder wav2vec2 checkpoint config',
)
parser.add_argument(
'--decoder_config_path',
default='facebook/s2t-small-mustc-en-fr-st',
type=str,
help='Path to hf decoder s2t checkpoint config',
)
parser.add_argument('--vocab_size', default=1_0224, type=int, help='Vocab size of decoder')
parser.add_argument('--num_decoder_layers', default=7, type=int, help='Number of decoder layers')
__a :Tuple = parser.parse_args()
convert_wavaveca_checkpoint(
args.checkpoint_path,
args.pytorch_dump_folder_path,
args.dict_path,
encoder_config_path=args.encoder_config_path,
decoder_config_path=args.decoder_config_path,
vocab_size=args.vocab_size,
num_decoder_layers=args.num_decoder_layers,
) | 86 | 1 |
from typing import Any, Callable, Dict, List, Optional, Union
import torch
from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer
from diffusers import (
AutoencoderKL,
DDIMScheduler,
DiffusionPipeline,
LMSDiscreteScheduler,
PNDMScheduler,
StableDiffusionPipeline,
UNetaDConditionModel,
)
from diffusers.pipelines.stable_diffusion import StableDiffusionPipelineOutput
from diffusers.pipelines.stable_diffusion.safety_checker import StableDiffusionSafetyChecker
__a :int = 'CompVis/stable-diffusion-v1-1'
__a :Optional[int] = 'CompVis/stable-diffusion-v1-2'
__a :List[Any] = 'CompVis/stable-diffusion-v1-3'
__a :int = 'CompVis/stable-diffusion-v1-4'
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Optional[Any] , UpperCAmelCase : AutoencoderKL , UpperCAmelCase : CLIPTextModel , UpperCAmelCase : CLIPTokenizer , UpperCAmelCase : UNetaDConditionModel , UpperCAmelCase : Union[DDIMScheduler, PNDMScheduler, LMSDiscreteScheduler] , UpperCAmelCase : StableDiffusionSafetyChecker , UpperCAmelCase : CLIPImageProcessor , UpperCAmelCase : bool = True , ):
super()._init_()
A_ = StableDiffusionPipeline.from_pretrained(UpperCAmelCase )
A_ = StableDiffusionPipeline.from_pretrained(UpperCAmelCase )
A_ = StableDiffusionPipeline.from_pretrained(UpperCAmelCase )
A_ = StableDiffusionPipeline(
vae=UpperCAmelCase , text_encoder=UpperCAmelCase , tokenizer=UpperCAmelCase , unet=UpperCAmelCase , scheduler=UpperCAmelCase , safety_checker=UpperCAmelCase , feature_extractor=UpperCAmelCase , requires_safety_checker=UpperCAmelCase , )
self.register_modules(pipelinea=self.pipea , pipelinea=self.pipea , pipelinea=self.pipea , pipelinea=self.pipea )
@property
def __A ( self : Union[str, Any] ):
return {k: getattr(self , UpperCAmelCase ) for k in self.config.keys() if not k.startswith("_" )}
def __A ( self : str , UpperCAmelCase : Optional[Union[str, int]] = "auto" ):
if slice_size == "auto":
# half the attention head size is usually a good trade-off between
# speed and memory
A_ = self.unet.config.attention_head_dim // 2
self.unet.set_attention_slice(UpperCAmelCase )
def __A ( self : List[Any] ):
self.enable_attention_slicing(UpperCAmelCase )
@torch.no_grad()
def __A ( self : Optional[Any] , UpperCAmelCase : Union[str, List[str]] , UpperCAmelCase : int = 512 , UpperCAmelCase : int = 512 , UpperCAmelCase : int = 50 , UpperCAmelCase : float = 7.5 , UpperCAmelCase : Optional[Union[str, List[str]]] = None , UpperCAmelCase : Optional[int] = 1 , UpperCAmelCase : float = 0.0 , UpperCAmelCase : Optional[torch.Generator] = None , UpperCAmelCase : Optional[torch.FloatTensor] = None , UpperCAmelCase : Optional[str] = "pil" , UpperCAmelCase : bool = True , UpperCAmelCase : Optional[Callable[[int, int, torch.FloatTensor], None]] = None , UpperCAmelCase : int = 1 , **UpperCAmelCase : Any , ):
return self.pipea(
prompt=UpperCAmelCase , height=UpperCAmelCase , width=UpperCAmelCase , num_inference_steps=UpperCAmelCase , guidance_scale=UpperCAmelCase , negative_prompt=UpperCAmelCase , num_images_per_prompt=UpperCAmelCase , eta=UpperCAmelCase , generator=UpperCAmelCase , latents=UpperCAmelCase , output_type=UpperCAmelCase , return_dict=UpperCAmelCase , callback=UpperCAmelCase , callback_steps=UpperCAmelCase , **UpperCAmelCase , )
@torch.no_grad()
def __A ( self : Optional[Any] , UpperCAmelCase : Union[str, List[str]] , UpperCAmelCase : int = 512 , UpperCAmelCase : int = 512 , UpperCAmelCase : int = 50 , UpperCAmelCase : float = 7.5 , UpperCAmelCase : Optional[Union[str, List[str]]] = None , UpperCAmelCase : Optional[int] = 1 , UpperCAmelCase : float = 0.0 , UpperCAmelCase : Optional[torch.Generator] = None , UpperCAmelCase : Optional[torch.FloatTensor] = None , UpperCAmelCase : Optional[str] = "pil" , UpperCAmelCase : bool = True , UpperCAmelCase : Optional[Callable[[int, int, torch.FloatTensor], None]] = None , UpperCAmelCase : int = 1 , **UpperCAmelCase : Dict , ):
return self.pipea(
prompt=UpperCAmelCase , height=UpperCAmelCase , width=UpperCAmelCase , num_inference_steps=UpperCAmelCase , guidance_scale=UpperCAmelCase , negative_prompt=UpperCAmelCase , num_images_per_prompt=UpperCAmelCase , eta=UpperCAmelCase , generator=UpperCAmelCase , latents=UpperCAmelCase , output_type=UpperCAmelCase , return_dict=UpperCAmelCase , callback=UpperCAmelCase , callback_steps=UpperCAmelCase , **UpperCAmelCase , )
@torch.no_grad()
def __A ( self : Dict , UpperCAmelCase : Union[str, List[str]] , UpperCAmelCase : int = 512 , UpperCAmelCase : int = 512 , UpperCAmelCase : int = 50 , UpperCAmelCase : float = 7.5 , UpperCAmelCase : Optional[Union[str, List[str]]] = None , UpperCAmelCase : Optional[int] = 1 , UpperCAmelCase : float = 0.0 , UpperCAmelCase : Optional[torch.Generator] = None , UpperCAmelCase : Optional[torch.FloatTensor] = None , UpperCAmelCase : Optional[str] = "pil" , UpperCAmelCase : bool = True , UpperCAmelCase : Optional[Callable[[int, int, torch.FloatTensor], None]] = None , UpperCAmelCase : int = 1 , **UpperCAmelCase : Tuple , ):
return self.pipea(
prompt=UpperCAmelCase , height=UpperCAmelCase , width=UpperCAmelCase , num_inference_steps=UpperCAmelCase , guidance_scale=UpperCAmelCase , negative_prompt=UpperCAmelCase , num_images_per_prompt=UpperCAmelCase , eta=UpperCAmelCase , generator=UpperCAmelCase , latents=UpperCAmelCase , output_type=UpperCAmelCase , return_dict=UpperCAmelCase , callback=UpperCAmelCase , callback_steps=UpperCAmelCase , **UpperCAmelCase , )
@torch.no_grad()
def __A ( self : List[str] , UpperCAmelCase : Union[str, List[str]] , UpperCAmelCase : int = 512 , UpperCAmelCase : int = 512 , UpperCAmelCase : int = 50 , UpperCAmelCase : float = 7.5 , UpperCAmelCase : Optional[Union[str, List[str]]] = None , UpperCAmelCase : Optional[int] = 1 , UpperCAmelCase : float = 0.0 , UpperCAmelCase : Optional[torch.Generator] = None , UpperCAmelCase : Optional[torch.FloatTensor] = None , UpperCAmelCase : Optional[str] = "pil" , UpperCAmelCase : bool = True , UpperCAmelCase : Optional[Callable[[int, int, torch.FloatTensor], None]] = None , UpperCAmelCase : int = 1 , **UpperCAmelCase : Tuple , ):
return self.pipea(
prompt=UpperCAmelCase , height=UpperCAmelCase , width=UpperCAmelCase , num_inference_steps=UpperCAmelCase , guidance_scale=UpperCAmelCase , negative_prompt=UpperCAmelCase , num_images_per_prompt=UpperCAmelCase , eta=UpperCAmelCase , generator=UpperCAmelCase , latents=UpperCAmelCase , output_type=UpperCAmelCase , return_dict=UpperCAmelCase , callback=UpperCAmelCase , callback_steps=UpperCAmelCase , **UpperCAmelCase , )
@torch.no_grad()
def __A ( self : str , UpperCAmelCase : Union[str, List[str]] , UpperCAmelCase : int = 512 , UpperCAmelCase : int = 512 , UpperCAmelCase : int = 50 , UpperCAmelCase : float = 7.5 , UpperCAmelCase : Optional[Union[str, List[str]]] = None , UpperCAmelCase : Optional[int] = 1 , UpperCAmelCase : float = 0.0 , UpperCAmelCase : Optional[torch.Generator] = None , UpperCAmelCase : Optional[torch.FloatTensor] = None , UpperCAmelCase : Optional[str] = "pil" , UpperCAmelCase : bool = True , UpperCAmelCase : Optional[Callable[[int, int, torch.FloatTensor], None]] = None , UpperCAmelCase : int = 1 , **UpperCAmelCase : Any , ):
A_ = "cuda" if torch.cuda.is_available() else "cpu"
self.to(UpperCAmelCase )
# Checks if the height and width are divisible by 8 or not
if height % 8 != 0 or width % 8 != 0:
raise ValueError(f'''`height` and `width` must be divisible by 8 but are {height} and {width}.''' )
# Get first result from Stable Diffusion Checkpoint v1.1
A_ = self.textaimg_sda_a(
prompt=UpperCAmelCase , height=UpperCAmelCase , width=UpperCAmelCase , num_inference_steps=UpperCAmelCase , guidance_scale=UpperCAmelCase , negative_prompt=UpperCAmelCase , num_images_per_prompt=UpperCAmelCase , eta=UpperCAmelCase , generator=UpperCAmelCase , latents=UpperCAmelCase , output_type=UpperCAmelCase , return_dict=UpperCAmelCase , callback=UpperCAmelCase , callback_steps=UpperCAmelCase , **UpperCAmelCase , )
# Get first result from Stable Diffusion Checkpoint v1.2
A_ = self.textaimg_sda_a(
prompt=UpperCAmelCase , height=UpperCAmelCase , width=UpperCAmelCase , num_inference_steps=UpperCAmelCase , guidance_scale=UpperCAmelCase , negative_prompt=UpperCAmelCase , num_images_per_prompt=UpperCAmelCase , eta=UpperCAmelCase , generator=UpperCAmelCase , latents=UpperCAmelCase , output_type=UpperCAmelCase , return_dict=UpperCAmelCase , callback=UpperCAmelCase , callback_steps=UpperCAmelCase , **UpperCAmelCase , )
# Get first result from Stable Diffusion Checkpoint v1.3
A_ = self.textaimg_sda_a(
prompt=UpperCAmelCase , height=UpperCAmelCase , width=UpperCAmelCase , num_inference_steps=UpperCAmelCase , guidance_scale=UpperCAmelCase , negative_prompt=UpperCAmelCase , num_images_per_prompt=UpperCAmelCase , eta=UpperCAmelCase , generator=UpperCAmelCase , latents=UpperCAmelCase , output_type=UpperCAmelCase , return_dict=UpperCAmelCase , callback=UpperCAmelCase , callback_steps=UpperCAmelCase , **UpperCAmelCase , )
# Get first result from Stable Diffusion Checkpoint v1.4
A_ = self.textaimg_sda_a(
prompt=UpperCAmelCase , height=UpperCAmelCase , width=UpperCAmelCase , num_inference_steps=UpperCAmelCase , guidance_scale=UpperCAmelCase , negative_prompt=UpperCAmelCase , num_images_per_prompt=UpperCAmelCase , eta=UpperCAmelCase , generator=UpperCAmelCase , latents=UpperCAmelCase , output_type=UpperCAmelCase , return_dict=UpperCAmelCase , callback=UpperCAmelCase , callback_steps=UpperCAmelCase , **UpperCAmelCase , )
# Get all result images into a single list and pass it via StableDiffusionPipelineOutput for final result
return StableDiffusionPipelineOutput([resa[0], resa[0], resa[0], resa[0]] ) | 86 |
from typing import Any, Dict, List, Union
from ..utils import add_end_docstrings, is_torch_available, is_vision_available, logging, requires_backends
from .base import PIPELINE_INIT_ARGS, Pipeline
if is_vision_available():
from ..image_utils import load_image
if is_torch_available():
import torch
from ..models.auto.modeling_auto import MODEL_FOR_OBJECT_DETECTION_MAPPING, MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING
__a :str = logging.get_logger(__name__)
__a :Any = Dict[str, Any]
__a :int = List[Prediction]
@add_end_docstrings(snake_case_ )
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Tuple , *UpperCAmelCase : Optional[int] , **UpperCAmelCase : Optional[Any] ):
super().__init__(*UpperCAmelCase , **UpperCAmelCase )
if self.framework == "tf":
raise ValueError(f'''The {self.__class__} is only available in PyTorch.''' )
requires_backends(self , "vision" )
self.check_model_type(
dict(MODEL_FOR_OBJECT_DETECTION_MAPPING.items() + MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING.items() ) )
def __A ( self : str , **UpperCAmelCase : str ):
A_ = {}
if "threshold" in kwargs:
A_ = kwargs["threshold"]
return {}, {}, postprocess_kwargs
def __call__( self : Union[str, Any] , *UpperCAmelCase : Union[str, Any] , **UpperCAmelCase : Optional[Any] ):
return super().__call__(*UpperCAmelCase , **UpperCAmelCase )
def __A ( self : str , UpperCAmelCase : Any ):
A_ = load_image(UpperCAmelCase )
A_ = torch.IntTensor([[image.height, image.width]] )
A_ = self.image_processor(images=[image] , return_tensors="pt" )
if self.tokenizer is not None:
A_ = self.tokenizer(text=inputs["words"] , boxes=inputs["boxes"] , return_tensors="pt" )
A_ = target_size
return inputs
def __A ( self : Optional[Any] , UpperCAmelCase : Optional[int] ):
A_ = model_inputs.pop("target_size" )
A_ = self.model(**UpperCAmelCase )
A_ = outputs.__class__({"target_size": target_size, **outputs} )
if self.tokenizer is not None:
A_ = model_inputs["bbox"]
return model_outputs
def __A ( self : str , UpperCAmelCase : Optional[int] , UpperCAmelCase : Union[str, Any]=0.9 ):
A_ = model_outputs["target_size"]
if self.tokenizer is not None:
# This is a LayoutLMForTokenClassification variant.
# The OCR got the boxes and the model classified the words.
A_ , A_ = target_size[0].tolist()
def unnormalize(UpperCAmelCase : Any ):
return self._get_bounding_box(
torch.Tensor(
[
(width * bbox[0] / 1000),
(height * bbox[1] / 1000),
(width * bbox[2] / 1000),
(height * bbox[3] / 1000),
] ) )
A_ , A_ = model_outputs["logits"].squeeze(0 ).softmax(dim=-1 ).max(dim=-1 )
A_ = [self.model.config.idalabel[prediction] for prediction in classes.tolist()]
A_ = [unnormalize(UpperCAmelCase ) for bbox in model_outputs["bbox"].squeeze(0 )]
A_ = ["score", "label", "box"]
A_ = [dict(zip(UpperCAmelCase , UpperCAmelCase ) ) for vals in zip(scores.tolist() , UpperCAmelCase , UpperCAmelCase ) if vals[0] > threshold]
else:
# This is a regular ForObjectDetectionModel
A_ = self.image_processor.post_process_object_detection(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase )
A_ = raw_annotations[0]
A_ = raw_annotation["scores"]
A_ = raw_annotation["labels"]
A_ = raw_annotation["boxes"]
A_ = scores.tolist()
A_ = [self.model.config.idalabel[label.item()] for label in labels]
A_ = [self._get_bounding_box(UpperCAmelCase ) for box in boxes]
# {"scores": [...], ...} --> [{"score":x, ...}, ...]
A_ = ["score", "label", "box"]
A_ = [
dict(zip(UpperCAmelCase , UpperCAmelCase ) )
for vals in zip(raw_annotation["scores"] , raw_annotation["labels"] , raw_annotation["boxes"] )
]
return annotation
def __A ( self : Tuple , UpperCAmelCase : "torch.Tensor" ):
if self.framework != "pt":
raise ValueError("The ObjectDetectionPipeline is only available in PyTorch." )
A_ , A_ , A_ , A_ = box.int().tolist()
A_ = {
"xmin": xmin,
"ymin": ymin,
"xmax": xmax,
"ymax": ymax,
}
return bbox | 86 | 1 |
from __future__ import absolute_import, division, print_function, unicode_literals
from torch import nn
from torch.nn import CrossEntropyLoss, MSELoss
from transformers import RobertaConfig
from transformers.file_utils import add_start_docstrings, add_start_docstrings_to_model_forward
from transformers.models.roberta.modeling_roberta import (
ROBERTA_INPUTS_DOCSTRING,
ROBERTA_START_DOCSTRING,
RobertaEmbeddings,
)
from .modeling_highway_bert import BertPreTrainedModel, DeeBertModel, HighwayException, entropy
@add_start_docstrings(
'The RoBERTa Model transformer with early exiting (DeeRoBERTa). ' , snake_case_ , )
class _a ( snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Union[str, Any] = RobertaConfig
_lowerCamelCase : Dict = 'roberta'
def __init__( self : Tuple , UpperCAmelCase : Tuple ):
super().__init__(UpperCAmelCase )
A_ = RobertaEmbeddings(UpperCAmelCase )
self.init_weights()
@add_start_docstrings(
'RoBERTa Model (with early exiting - DeeRoBERTa) with a classifier on top,\n also takes care of multi-layer training. ' , snake_case_ , )
class _a ( snake_case_ ):
"""simple docstring"""
_lowerCamelCase : str = RobertaConfig
_lowerCamelCase : List[str] = 'roberta'
def __init__( self : List[str] , UpperCAmelCase : List[Any] ):
super().__init__(UpperCAmelCase )
A_ = config.num_labels
A_ = config.num_hidden_layers
A_ = DeeRobertaModel(UpperCAmelCase )
A_ = nn.Dropout(config.hidden_dropout_prob )
A_ = nn.Linear(config.hidden_size , self.config.num_labels )
@add_start_docstrings_to_model_forward(UpperCAmelCase )
def __A ( self : Dict , UpperCAmelCase : Union[str, Any]=None , UpperCAmelCase : Dict=None , UpperCAmelCase : Dict=None , UpperCAmelCase : Optional[int]=None , UpperCAmelCase : Optional[Any]=None , UpperCAmelCase : int=None , UpperCAmelCase : Union[str, Any]=None , UpperCAmelCase : Optional[int]=-1 , UpperCAmelCase : Optional[int]=False , ):
A_ = self.num_layers
try:
A_ = self.roberta(
UpperCAmelCase , attention_mask=UpperCAmelCase , token_type_ids=UpperCAmelCase , position_ids=UpperCAmelCase , head_mask=UpperCAmelCase , inputs_embeds=UpperCAmelCase , )
A_ = outputs[1]
A_ = self.dropout(UpperCAmelCase )
A_ = self.classifier(UpperCAmelCase )
A_ = (logits,) + outputs[2:] # add hidden states and attention if they are here
except HighwayException as e:
A_ = e.message
A_ = e.exit_layer
A_ = outputs[0]
if not self.training:
A_ = entropy(UpperCAmelCase )
A_ = []
A_ = []
if labels is not None:
if self.num_labels == 1:
# We are doing regression
A_ = MSELoss()
A_ = loss_fct(logits.view(-1 ) , labels.view(-1 ) )
else:
A_ = CrossEntropyLoss()
A_ = loss_fct(logits.view(-1 , self.num_labels ) , labels.view(-1 ) )
# work with highway exits
A_ = []
for highway_exit in outputs[-1]:
A_ = highway_exit[0]
if not self.training:
highway_logits_all.append(UpperCAmelCase )
highway_entropy.append(highway_exit[2] )
if self.num_labels == 1:
# We are doing regression
A_ = MSELoss()
A_ = loss_fct(highway_logits.view(-1 ) , labels.view(-1 ) )
else:
A_ = CrossEntropyLoss()
A_ = loss_fct(highway_logits.view(-1 , self.num_labels ) , labels.view(-1 ) )
highway_losses.append(UpperCAmelCase )
if train_highway:
A_ = (sum(highway_losses[:-1] ),) + outputs
# exclude the final highway, of course
else:
A_ = (loss,) + outputs
if not self.training:
A_ = outputs + ((original_entropy, highway_entropy), exit_layer)
if output_layer >= 0:
A_ = (
(outputs[0],) + (highway_logits_all[output_layer],) + outputs[2:]
) # use the highway of the last layer
return outputs # (loss), logits, (hidden_states), (attentions), entropy | 86 |
import inspect
from typing import List, Optional, Tuple, Union
import numpy as np
import PIL
import torch
import torch.utils.checkpoint
from ...models import UNetaDModel, VQModel
from ...schedulers import (
DDIMScheduler,
DPMSolverMultistepScheduler,
EulerAncestralDiscreteScheduler,
EulerDiscreteScheduler,
LMSDiscreteScheduler,
PNDMScheduler,
)
from ...utils import PIL_INTERPOLATION, randn_tensor
from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
def __snake_case ( __UpperCamelCase : Dict ):
"""simple docstring"""
A_ , A_ = image.size
A_ , A_ = (x - x % 32 for x in (w, h)) # resize to integer multiple of 32
A_ = image.resize((w, h) ,resample=PIL_INTERPOLATION["lanczos"] )
A_ = np.array(__UpperCamelCase ).astype(np.floataa ) / 255.0
A_ = image[None].transpose(0 ,3 ,1 ,2 )
A_ = torch.from_numpy(__UpperCamelCase )
return 2.0 * image - 1.0
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Union[str, Any] , UpperCAmelCase : VQModel , UpperCAmelCase : UNetaDModel , UpperCAmelCase : Union[
DDIMScheduler,
PNDMScheduler,
LMSDiscreteScheduler,
EulerDiscreteScheduler,
EulerAncestralDiscreteScheduler,
DPMSolverMultistepScheduler,
] , ):
super().__init__()
self.register_modules(vqvae=UpperCAmelCase , unet=UpperCAmelCase , scheduler=UpperCAmelCase )
@torch.no_grad()
def __call__( self : int , UpperCAmelCase : Union[torch.Tensor, PIL.Image.Image] = None , UpperCAmelCase : Optional[int] = 1 , UpperCAmelCase : Optional[int] = 100 , UpperCAmelCase : Optional[float] = 0.0 , UpperCAmelCase : Optional[Union[torch.Generator, List[torch.Generator]]] = None , UpperCAmelCase : Optional[str] = "pil" , UpperCAmelCase : bool = True , ):
if isinstance(UpperCAmelCase , PIL.Image.Image ):
A_ = 1
elif isinstance(UpperCAmelCase , torch.Tensor ):
A_ = image.shape[0]
else:
raise ValueError(f'''`image` has to be of type `PIL.Image.Image` or `torch.Tensor` but is {type(UpperCAmelCase )}''' )
if isinstance(UpperCAmelCase , PIL.Image.Image ):
A_ = preprocess(UpperCAmelCase )
A_ , A_ = image.shape[-2:]
# in_channels should be 6: 3 for latents, 3 for low resolution image
A_ = (batch_size, self.unet.config.in_channels // 2, height, width)
A_ = next(self.unet.parameters() ).dtype
A_ = randn_tensor(UpperCAmelCase , generator=UpperCAmelCase , device=self.device , dtype=UpperCAmelCase )
A_ = image.to(device=self.device , dtype=UpperCAmelCase )
# set timesteps and move to the correct device
self.scheduler.set_timesteps(UpperCAmelCase , device=self.device )
A_ = self.scheduler.timesteps
# scale the initial noise by the standard deviation required by the scheduler
A_ = latents * self.scheduler.init_noise_sigma
# prepare extra kwargs for the scheduler step, since not all schedulers have the same signature.
# eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
# eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
# and should be between [0, 1]
A_ = "eta" in set(inspect.signature(self.scheduler.step ).parameters.keys() )
A_ = {}
if accepts_eta:
A_ = eta
for t in self.progress_bar(UpperCAmelCase ):
# concat latents and low resolution image in the channel dimension.
A_ = torch.cat([latents, image] , dim=1 )
A_ = self.scheduler.scale_model_input(UpperCAmelCase , UpperCAmelCase )
# predict the noise residual
A_ = self.unet(UpperCAmelCase , UpperCAmelCase ).sample
# compute the previous noisy sample x_t -> x_t-1
A_ = self.scheduler.step(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase , **UpperCAmelCase ).prev_sample
# decode the image latents with the VQVAE
A_ = self.vqvae.decode(UpperCAmelCase ).sample
A_ = torch.clamp(UpperCAmelCase , -1.0 , 1.0 )
A_ = image / 2 + 0.5
A_ = image.cpu().permute(0 , 2 , 3 , 1 ).numpy()
if output_type == "pil":
A_ = self.numpy_to_pil(UpperCAmelCase )
if not return_dict:
return (image,)
return ImagePipelineOutput(images=UpperCAmelCase ) | 86 | 1 |
import json
import os
import unittest
from transformers.models.roc_bert.tokenization_roc_bert import (
VOCAB_FILES_NAMES,
RoCBertBasicTokenizer,
RoCBertTokenizer,
RoCBertWordpieceTokenizer,
_is_control,
_is_punctuation,
_is_whitespace,
)
from transformers.testing_utils import require_tokenizers, slow
from ...test_tokenization_common import TokenizerTesterMixin, filter_non_english
@require_tokenizers
class _a ( snake_case_ , unittest.TestCase ):
"""simple docstring"""
_lowerCamelCase : Dict = RoCBertTokenizer
_lowerCamelCase : str = None
_lowerCamelCase : Dict = False
_lowerCamelCase : Tuple = True
_lowerCamelCase : List[str] = filter_non_english
def __A ( self : List[Any] ):
super().setUp()
A_ = ["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]", "你", "好", "是", "谁", "a", "b", "c", "d"]
A_ = {}
A_ = {}
for i, value in enumerate(UpperCAmelCase ):
A_ = i
A_ = i
A_ = os.path.join(self.tmpdirname , VOCAB_FILES_NAMES["vocab_file"] )
A_ = os.path.join(self.tmpdirname , VOCAB_FILES_NAMES["word_shape_file"] )
A_ = os.path.join(self.tmpdirname , VOCAB_FILES_NAMES["word_pronunciation_file"] )
with open(self.vocab_file , "w" , encoding="utf-8" ) as vocab_writer:
vocab_writer.write("".join([x + "\n" for x in vocab_tokens] ) )
with open(self.word_shape_file , "w" , encoding="utf-8" ) as word_shape_writer:
json.dump(UpperCAmelCase , UpperCAmelCase , ensure_ascii=UpperCAmelCase )
with open(self.word_pronunciation_file , "w" , encoding="utf-8" ) as word_pronunciation_writer:
json.dump(UpperCAmelCase , UpperCAmelCase , ensure_ascii=UpperCAmelCase )
def __A ( self : Dict ):
A_ = self.tokenizer_class(self.vocab_file , self.word_shape_file , self.word_pronunciation_file )
A_ = tokenizer.tokenize("你好[SEP]你是谁" )
self.assertListEqual(UpperCAmelCase , ["你", "好", "[SEP]", "你", "是", "谁"] )
self.assertListEqual(tokenizer.convert_tokens_to_ids(UpperCAmelCase ) , [5, 6, 2, 5, 7, 8] )
self.assertListEqual(tokenizer.convert_tokens_to_shape_ids(UpperCAmelCase ) , [5, 6, 2, 5, 7, 8] )
self.assertListEqual(tokenizer.convert_tokens_to_pronunciation_ids(UpperCAmelCase ) , [5, 6, 2, 5, 7, 8] )
def __A ( self : Optional[Any] ):
A_ = RoCBertBasicTokenizer()
self.assertListEqual(tokenizer.tokenize("ah\u535A\u63A8zz" ) , ["ah", "\u535A", "\u63A8", "zz"] )
def __A ( self : Optional[int] ):
A_ = RoCBertBasicTokenizer(do_lower_case=UpperCAmelCase )
self.assertListEqual(
tokenizer.tokenize(" \tHeLLo!how \n Are yoU? " ) , ["hello", "!", "how", "are", "you", "?"] )
self.assertListEqual(tokenizer.tokenize("H\u00E9llo" ) , ["hello"] )
def __A ( self : Dict ):
A_ = RoCBertBasicTokenizer(do_lower_case=UpperCAmelCase , strip_accents=UpperCAmelCase )
self.assertListEqual(
tokenizer.tokenize(" \tHäLLo!how \n Are yoU? " ) , ["hällo", "!", "how", "are", "you", "?"] )
self.assertListEqual(tokenizer.tokenize("H\u00E9llo" ) , ["h\u00E9llo"] )
def __A ( self : Optional[Any] ):
A_ = RoCBertBasicTokenizer(do_lower_case=UpperCAmelCase , strip_accents=UpperCAmelCase )
self.assertListEqual(
tokenizer.tokenize(" \tHäLLo!how \n Are yoU? " ) , ["hallo", "!", "how", "are", "you", "?"] )
self.assertListEqual(tokenizer.tokenize("H\u00E9llo" ) , ["hello"] )
def __A ( self : Dict ):
A_ = RoCBertBasicTokenizer(do_lower_case=UpperCAmelCase )
self.assertListEqual(
tokenizer.tokenize(" \tHäLLo!how \n Are yoU? " ) , ["hallo", "!", "how", "are", "you", "?"] )
self.assertListEqual(tokenizer.tokenize("H\u00E9llo" ) , ["hello"] )
def __A ( self : List[str] ):
A_ = RoCBertBasicTokenizer(do_lower_case=UpperCAmelCase )
self.assertListEqual(
tokenizer.tokenize(" \tHeLLo!how \n Are yoU? " ) , ["HeLLo", "!", "how", "Are", "yoU", "?"] )
def __A ( self : int ):
A_ = RoCBertBasicTokenizer(do_lower_case=UpperCAmelCase , strip_accents=UpperCAmelCase )
self.assertListEqual(
tokenizer.tokenize(" \tHäLLo!how \n Are yoU? " ) , ["HäLLo", "!", "how", "Are", "yoU", "?"] )
def __A ( self : Optional[Any] ):
A_ = RoCBertBasicTokenizer(do_lower_case=UpperCAmelCase , strip_accents=UpperCAmelCase )
self.assertListEqual(
tokenizer.tokenize(" \tHäLLo!how \n Are yoU? " ) , ["HaLLo", "!", "how", "Are", "yoU", "?"] )
def __A ( self : List[str] ):
A_ = RoCBertBasicTokenizer(do_lower_case=UpperCAmelCase , never_split=["[UNK]"] )
self.assertListEqual(
tokenizer.tokenize(" \tHeLLo!how \n Are yoU? [UNK]" ) , ["HeLLo", "!", "how", "Are", "yoU", "?", "[UNK]"] )
def __A ( self : Optional[Any] ):
A_ = ["[UNK]", "[CLS]", "[SEP]", "want", "##want", "##ed", "wa", "un", "runn", "##ing"]
A_ = {}
for i, token in enumerate(UpperCAmelCase ):
A_ = i
A_ = RoCBertWordpieceTokenizer(vocab=UpperCAmelCase , unk_token="[UNK]" )
self.assertListEqual(tokenizer.tokenize("" ) , [] )
self.assertListEqual(tokenizer.tokenize("unwanted running" ) , ["un", "##want", "##ed", "runn", "##ing"] )
self.assertListEqual(tokenizer.tokenize("unwantedX running" ) , ["[UNK]", "runn", "##ing"] )
def __A ( self : Dict ):
self.assertTrue(_is_whitespace(" " ) )
self.assertTrue(_is_whitespace("\t" ) )
self.assertTrue(_is_whitespace("\r" ) )
self.assertTrue(_is_whitespace("\n" ) )
self.assertTrue(_is_whitespace("\u00A0" ) )
self.assertFalse(_is_whitespace("A" ) )
self.assertFalse(_is_whitespace("-" ) )
def __A ( self : Any ):
self.assertTrue(_is_control("\u0005" ) )
self.assertFalse(_is_control("A" ) )
self.assertFalse(_is_control(" " ) )
self.assertFalse(_is_control("\t" ) )
self.assertFalse(_is_control("\r" ) )
def __A ( self : Dict ):
self.assertTrue(_is_punctuation("-" ) )
self.assertTrue(_is_punctuation("$" ) )
self.assertTrue(_is_punctuation("`" ) )
self.assertTrue(_is_punctuation("." ) )
self.assertFalse(_is_punctuation("A" ) )
self.assertFalse(_is_punctuation(" " ) )
def __A ( self : List[str] ):
A_ = self.get_tokenizer()
# Example taken from the issue https://github.com/huggingface/tokenizers/issues/340
self.assertListEqual([tokenizer.tokenize(UpperCAmelCase ) for t in ["Test", "\xad", "test"]] , [["[UNK]"], [], ["[UNK]"]] )
if self.test_rust_tokenizer:
A_ = self.get_rust_tokenizer()
self.assertListEqual(
[rust_tokenizer.tokenize(UpperCAmelCase ) for t in ["Test", "\xad", "test"]] , [["[UNK]"], [], ["[UNK]"]] )
def __A ( self : List[str] ):
for tokenizer, pretrained_name, kwargs in self.tokenizers_list:
with self.subTest(f'''{tokenizer.__class__.__name__} ({pretrained_name})''' ):
A_ = self.rust_tokenizer_class.from_pretrained(UpperCAmelCase , **UpperCAmelCase )
A_ = f'''A, naïve {tokenizer_r.mask_token} AllenNLP sentence.'''
A_ = tokenizer_r.encode_plus(
UpperCAmelCase , return_attention_mask=UpperCAmelCase , return_token_type_ids=UpperCAmelCase , return_offsets_mapping=UpperCAmelCase , add_special_tokens=UpperCAmelCase , )
A_ = tokenizer_r.do_lower_case if hasattr(UpperCAmelCase , "do_lower_case" ) else False
A_ = (
[
((0, 0), tokenizer_r.cls_token),
((0, 1), "A"),
((1, 2), ","),
((3, 5), "na"),
((5, 6), "##ï"),
((6, 8), "##ve"),
((9, 15), tokenizer_r.mask_token),
((16, 21), "Allen"),
((21, 23), "##NL"),
((23, 24), "##P"),
((25, 33), "sentence"),
((33, 34), "."),
((0, 0), tokenizer_r.sep_token),
]
if not do_lower_case
else [
((0, 0), tokenizer_r.cls_token),
((0, 1), "a"),
((1, 2), ","),
((3, 8), "naive"),
((9, 15), tokenizer_r.mask_token),
((16, 21), "allen"),
((21, 23), "##nl"),
((23, 24), "##p"),
((25, 33), "sentence"),
((33, 34), "."),
((0, 0), tokenizer_r.sep_token),
]
)
self.assertEqual(
[e[1] for e in expected_results] , tokenizer_r.convert_ids_to_tokens(tokens["input_ids"] ) )
self.assertEqual([e[0] for e in expected_results] , tokens["offset_mapping"] )
def __A ( self : Tuple ):
A_ = ["的", "人", "有"]
A_ = "".join(UpperCAmelCase )
for tokenizer, pretrained_name, kwargs in self.tokenizers_list:
with self.subTest(f'''{tokenizer.__class__.__name__} ({pretrained_name})''' ):
A_ = True
A_ = self.tokenizer_class.from_pretrained(UpperCAmelCase , **UpperCAmelCase )
A_ = self.rust_tokenizer_class.from_pretrained(UpperCAmelCase , **UpperCAmelCase )
A_ = tokenizer_p.encode(UpperCAmelCase , add_special_tokens=UpperCAmelCase )
A_ = tokenizer_r.encode(UpperCAmelCase , add_special_tokens=UpperCAmelCase )
A_ = tokenizer_r.convert_ids_to_tokens(UpperCAmelCase )
A_ = tokenizer_p.convert_ids_to_tokens(UpperCAmelCase )
# it is expected that each Chinese character is not preceded by "##"
self.assertListEqual(UpperCAmelCase , UpperCAmelCase )
self.assertListEqual(UpperCAmelCase , UpperCAmelCase )
A_ = False
A_ = self.rust_tokenizer_class.from_pretrained(UpperCAmelCase , **UpperCAmelCase )
A_ = self.tokenizer_class.from_pretrained(UpperCAmelCase , **UpperCAmelCase )
A_ = tokenizer_r.encode(UpperCAmelCase , add_special_tokens=UpperCAmelCase )
A_ = tokenizer_p.encode(UpperCAmelCase , add_special_tokens=UpperCAmelCase )
A_ = tokenizer_r.convert_ids_to_tokens(UpperCAmelCase )
A_ = tokenizer_p.convert_ids_to_tokens(UpperCAmelCase )
# it is expected that only the first Chinese character is not preceded by "##".
A_ = [
f'''##{token}''' if idx != 0 else token for idx, token in enumerate(UpperCAmelCase )
]
self.assertListEqual(UpperCAmelCase , UpperCAmelCase )
self.assertListEqual(UpperCAmelCase , UpperCAmelCase )
@slow
def __A ( self : List[str] ):
A_ = self.tokenizer_class(self.vocab_file , self.word_shape_file , self.word_pronunciation_file )
A_ = tokenizer.encode("你好" , add_special_tokens=UpperCAmelCase )
A_ = tokenizer.encode("你是谁" , add_special_tokens=UpperCAmelCase )
A_ = tokenizer.build_inputs_with_special_tokens(UpperCAmelCase )
A_ = tokenizer.build_inputs_with_special_tokens(UpperCAmelCase , UpperCAmelCase )
assert encoded_sentence == [1] + text + [2]
assert encoded_pair == [1] + text + [2] + text_a + [2]
def __A ( self : Any ):
A_ = self.get_tokenizers(do_lower_case=UpperCAmelCase )
for tokenizer in tokenizers:
with self.subTest(f'''{tokenizer.__class__.__name__}''' ):
A_ = "你好，你是谁"
A_ = tokenizer.tokenize(UpperCAmelCase )
A_ = tokenizer.convert_tokens_to_ids(UpperCAmelCase )
A_ = tokenizer.convert_tokens_to_shape_ids(UpperCAmelCase )
A_ = tokenizer.convert_tokens_to_pronunciation_ids(UpperCAmelCase )
A_ = tokenizer.prepare_for_model(
UpperCAmelCase , UpperCAmelCase , UpperCAmelCase , add_special_tokens=UpperCAmelCase )
A_ = tokenizer.encode_plus(UpperCAmelCase , add_special_tokens=UpperCAmelCase )
self.assertEqual(UpperCAmelCase , UpperCAmelCase ) | 86 |
__a :Optional[int] = [sum(int(c, 10) ** 2 for c in i.__str__()) for i in range(10_0000)]
def __snake_case ( __UpperCamelCase : int ):
"""simple docstring"""
A_ = 0
while number:
# Increased Speed Slightly by checking every 5 digits together.
sum_of_digits_squared += DIGITS_SQUARED[number % 10_0000]
number //= 10_0000
return sum_of_digits_squared
# There are 2 Chains made,
# One ends with 89 with the chain member 58 being the one which when declared first,
# there will be the least number of iterations for all the members to be checked.
# The other one ends with 1 and has only one element 1.
# So 58 and 1 are chosen to be declared at the starting.
# Changed dictionary to an array to quicken the solution
__a :list[bool | None] = [None] * 1000_0000
__a :Optional[Any] = True
__a :List[Any] = False
def __snake_case ( __UpperCamelCase : int ):
"""simple docstring"""
if CHAINS[number - 1] is not None:
return CHAINS[number - 1] # type: ignore
A_ = chain(next_number(__UpperCamelCase ) )
A_ = number_chain
while number < 1000_0000:
A_ = number_chain
number *= 10
return number_chain
def __snake_case ( __UpperCamelCase : int = 1000_0000 ):
"""simple docstring"""
for i in range(1 ,__UpperCamelCase ):
if CHAINS[i] is None:
chain(i + 1 )
return CHAINS[:number].count(__UpperCamelCase )
if __name__ == "__main__":
import doctest
doctest.testmod()
print(F"{solution() = }") | 86 | 1 |
from typing import TYPE_CHECKING
from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_torch_available, is_vision_available
__a :Any = {
'configuration_chinese_clip': [
'CHINESE_CLIP_PRETRAINED_CONFIG_ARCHIVE_MAP',
'ChineseCLIPConfig',
'ChineseCLIPOnnxConfig',
'ChineseCLIPTextConfig',
'ChineseCLIPVisionConfig',
],
'processing_chinese_clip': ['ChineseCLIPProcessor'],
}
try:
if not is_vision_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Tuple = ['ChineseCLIPFeatureExtractor']
__a :List[Any] = ['ChineseCLIPImageProcessor']
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Tuple = [
'CHINESE_CLIP_PRETRAINED_MODEL_ARCHIVE_LIST',
'ChineseCLIPModel',
'ChineseCLIPPreTrainedModel',
'ChineseCLIPTextModel',
'ChineseCLIPVisionModel',
]
if TYPE_CHECKING:
from .configuration_chinese_clip import (
CHINESE_CLIP_PRETRAINED_CONFIG_ARCHIVE_MAP,
ChineseCLIPConfig,
ChineseCLIPOnnxConfig,
ChineseCLIPTextConfig,
ChineseCLIPVisionConfig,
)
from .processing_chinese_clip import ChineseCLIPProcessor
try:
if not is_vision_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .feature_extraction_chinese_clip import ChineseCLIPFeatureExtractor, ChineseCLIPImageProcessor
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_chinese_clip import (
CHINESE_CLIP_PRETRAINED_MODEL_ARCHIVE_LIST,
ChineseCLIPModel,
ChineseCLIPPreTrainedModel,
ChineseCLIPTextModel,
ChineseCLIPVisionModel,
)
else:
import sys
__a :List[Any] = _LazyModule(__name__, globals()['__file__'], _import_structure, module_spec=__spec__) | 86 |
from typing import TYPE_CHECKING
from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_tf_available, is_torch_available
__a :List[Any] = {
'configuration_tapas': ['TAPAS_PRETRAINED_CONFIG_ARCHIVE_MAP', 'TapasConfig'],
'tokenization_tapas': ['TapasTokenizer'],
}
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Any = [
'TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST',
'TapasForMaskedLM',
'TapasForQuestionAnswering',
'TapasForSequenceClassification',
'TapasModel',
'TapasPreTrainedModel',
'load_tf_weights_in_tapas',
]
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Dict = [
'TF_TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST',
'TFTapasForMaskedLM',
'TFTapasForQuestionAnswering',
'TFTapasForSequenceClassification',
'TFTapasModel',
'TFTapasPreTrainedModel',
]
if TYPE_CHECKING:
from .configuration_tapas import TAPAS_PRETRAINED_CONFIG_ARCHIVE_MAP, TapasConfig
from .tokenization_tapas import TapasTokenizer
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tapas import (
TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST,
TapasForMaskedLM,
TapasForQuestionAnswering,
TapasForSequenceClassification,
TapasModel,
TapasPreTrainedModel,
load_tf_weights_in_tapas,
)
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tf_tapas import (
TF_TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST,
TFTapasForMaskedLM,
TFTapasForQuestionAnswering,
TFTapasForSequenceClassification,
TFTapasModel,
TFTapasPreTrainedModel,
)
else:
import sys
__a :str = _LazyModule(__name__, globals()['__file__'], _import_structure, module_spec=__spec__) | 86 | 1 |
def __snake_case ( __UpperCamelCase : int ):
"""simple docstring"""
if num <= 0:
raise ValueError("Input must be a positive integer" )
A_ = [True] * (num + 1)
A_ = 2
while p * p <= num:
if primes[p]:
for i in range(p * p ,num + 1 ,__UpperCamelCase ):
A_ = False
p += 1
return [prime for prime in range(2 ,num + 1 ) if primes[prime]]
if __name__ == "__main__":
import doctest
doctest.testmod()
__a :List[Any] = int(input('Enter a positive integer: ').strip())
print(prime_sieve_eratosthenes(user_num)) | 86 |
# Lint as: python3
import sys
from collections.abc import Mapping
from typing import TYPE_CHECKING, Dict, Optional
import numpy as np
import pyarrow as pa
from .. import config
from ..utils.logging import get_logger
from ..utils.py_utils import map_nested
from .formatting import TensorFormatter
if TYPE_CHECKING:
import jax
import jaxlib
__a :List[Any] = get_logger()
__a :Optional[dict] = None
class _a ( TensorFormatter[Mapping, 'jax.Array', Mapping] ):
"""simple docstring"""
def __init__( self : str , UpperCAmelCase : int=None , UpperCAmelCase : List[str]=None , **UpperCAmelCase : List[Any] ):
super().__init__(features=UpperCAmelCase )
import jax
from jaxlib.xla_client import Device
if isinstance(UpperCAmelCase , UpperCAmelCase ):
raise ValueError(
f'''Expected {device} to be a `str` not {type(UpperCAmelCase )}, as `jaxlib.xla_extension.Device` '''
"is not serializable neither with `pickle` nor with `dill`. Instead you can surround "
"the device with `str()` to get its string identifier that will be internally mapped "
"to the actual `jaxlib.xla_extension.Device`." )
A_ = device if isinstance(UpperCAmelCase , UpperCAmelCase ) else str(jax.devices()[0] )
# using global variable since `jaxlib.xla_extension.Device` is not serializable neither
# with `pickle` nor with `dill`, so we need to use a global variable instead
global DEVICE_MAPPING
if DEVICE_MAPPING is None:
A_ = self._map_devices_to_str()
if self.device not in list(DEVICE_MAPPING.keys() ):
logger.warning(
f'''Device with string identifier {self.device} not listed among the available '''
f'''devices: {list(DEVICE_MAPPING.keys() )}, so falling back to the default '''
f'''device: {str(jax.devices()[0] )}.''' )
A_ = str(jax.devices()[0] )
A_ = jnp_array_kwargs
@staticmethod
def __A ( ):
import jax
return {str(UpperCAmelCase ): device for device in jax.devices()}
def __A ( self : Optional[int] , UpperCAmelCase : int ):
import jax
import jax.numpy as jnp
if isinstance(UpperCAmelCase , UpperCAmelCase ) and column:
if all(
isinstance(UpperCAmelCase , jax.Array ) and x.shape == column[0].shape and x.dtype == column[0].dtype for x in column ):
return jnp.stack(UpperCAmelCase , axis=0 )
return column
def __A ( self : List[str] , UpperCAmelCase : str ):
import jax
import jax.numpy as jnp
if isinstance(UpperCAmelCase , (str, bytes, type(UpperCAmelCase )) ):
return value
elif isinstance(UpperCAmelCase , (np.character, np.ndarray) ) and np.issubdtype(value.dtype , np.character ):
return value.tolist()
A_ = {}
if isinstance(UpperCAmelCase , (np.number, np.ndarray) ) and np.issubdtype(value.dtype , np.integer ):
# the default int precision depends on the jax config
# see https://jax.readthedocs.io/en/latest/notebooks/Common_Gotchas_in_JAX.html#double-64bit-precision
if jax.config.jax_enable_xaa:
A_ = {"dtype": jnp.intaa}
else:
A_ = {"dtype": jnp.intaa}
elif isinstance(UpperCAmelCase , (np.number, np.ndarray) ) and np.issubdtype(value.dtype , np.floating ):
A_ = {"dtype": jnp.floataa}
elif config.PIL_AVAILABLE and "PIL" in sys.modules:
import PIL.Image
if isinstance(UpperCAmelCase , PIL.Image.Image ):
A_ = np.asarray(UpperCAmelCase )
# using global variable since `jaxlib.xla_extension.Device` is not serializable neither
# with `pickle` nor with `dill`, so we need to use a global variable instead
global DEVICE_MAPPING
if DEVICE_MAPPING is None:
A_ = self._map_devices_to_str()
with jax.default_device(DEVICE_MAPPING[self.device] ):
# calling jnp.array on a np.ndarray does copy the data
# see https://github.com/google/jax/issues/4486
return jnp.array(UpperCAmelCase , **{**default_dtype, **self.jnp_array_kwargs} )
def __A ( self : Any , UpperCAmelCase : Dict ):
import jax
# support for torch, tf, jax etc.
if config.TORCH_AVAILABLE and "torch" in sys.modules:
import torch
if isinstance(UpperCAmelCase , torch.Tensor ):
return self._tensorize(data_struct.detach().cpu().numpy()[()] )
if hasattr(UpperCAmelCase , "__array__" ) and not isinstance(UpperCAmelCase , jax.Array ):
A_ = data_struct.__array__()
# support for nested types like struct of list of struct
if isinstance(UpperCAmelCase , np.ndarray ):
if data_struct.dtype == object: # jax arrays cannot be instantied from an array of objects
return self._consolidate([self.recursive_tensorize(UpperCAmelCase ) for substruct in data_struct] )
elif isinstance(UpperCAmelCase , (list, tuple) ):
return self._consolidate([self.recursive_tensorize(UpperCAmelCase ) for substruct in data_struct] )
return self._tensorize(UpperCAmelCase )
def __A ( self : Tuple , UpperCAmelCase : dict ):
return map_nested(self._recursive_tensorize , UpperCAmelCase , map_list=UpperCAmelCase )
def __A ( self : Dict , UpperCAmelCase : pa.Table ):
A_ = self.numpy_arrow_extractor().extract_row(UpperCAmelCase )
A_ = self.python_features_decoder.decode_row(UpperCAmelCase )
return self.recursive_tensorize(UpperCAmelCase )
def __A ( self : Any , UpperCAmelCase : pa.Table ):
A_ = self.numpy_arrow_extractor().extract_column(UpperCAmelCase )
A_ = self.python_features_decoder.decode_column(UpperCAmelCase , pa_table.column_names[0] )
A_ = self.recursive_tensorize(UpperCAmelCase )
A_ = self._consolidate(UpperCAmelCase )
return column
def __A ( self : Dict , UpperCAmelCase : pa.Table ):
A_ = self.numpy_arrow_extractor().extract_batch(UpperCAmelCase )
A_ = self.python_features_decoder.decode_batch(UpperCAmelCase )
A_ = self.recursive_tensorize(UpperCAmelCase )
for column_name in batch:
A_ = self._consolidate(batch[column_name] )
return batch | 86 | 1 |
import argparse
import os
from transformers.utils import direct_transformers_import
# All paths are set with the intent you should run this script from the root of the repo with the command
# python utils/check_task_guides.py
__a :Optional[Any] = 'src/transformers'
__a :Tuple = 'docs/source/en/tasks'
def __snake_case ( __UpperCamelCase : List[Any] ,__UpperCamelCase : Dict ,__UpperCamelCase : int ):
"""simple docstring"""
with open(__UpperCamelCase ,"r" ,encoding="utf-8" ,newline="\n" ) as f:
A_ = f.readlines()
# Find the start prompt.
A_ = 0
while not lines[start_index].startswith(__UpperCamelCase ):
start_index += 1
start_index += 1
A_ = start_index
while not lines[end_index].startswith(__UpperCamelCase ):
end_index += 1
end_index -= 1
while len(lines[start_index] ) <= 1:
start_index += 1
while len(lines[end_index] ) <= 1:
end_index -= 1
end_index += 1
return "".join(lines[start_index:end_index] ), start_index, end_index, lines
# This is to make sure the transformers module imported is the one in the repo.
__a :List[str] = direct_transformers_import(TRANSFORMERS_PATH)
__a :Optional[Any] = {
'asr.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_CTC_MAPPING_NAMES,
'audio_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_AUDIO_CLASSIFICATION_MAPPING_NAMES,
'language_modeling.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_CAUSAL_LM_MAPPING_NAMES,
'image_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_IMAGE_CLASSIFICATION_MAPPING_NAMES,
'masked_language_modeling.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_MASKED_LM_MAPPING_NAMES,
'multiple_choice.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_MULTIPLE_CHOICE_MAPPING_NAMES,
'object_detection.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_OBJECT_DETECTION_MAPPING_NAMES,
'question_answering.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_QUESTION_ANSWERING_MAPPING_NAMES,
'semantic_segmentation.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEMANTIC_SEGMENTATION_MAPPING_NAMES,
'sequence_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING_NAMES,
'summarization.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES,
'token_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING_NAMES,
'translation.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES,
'video_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_VIDEO_CLASSIFICATION_MAPPING_NAMES,
'document_question_answering.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_DOCUMENT_QUESTION_ANSWERING_MAPPING_NAMES,
'monocular_depth_estimation.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_DEPTH_ESTIMATION_MAPPING_NAMES,
}
# This list contains model types used in some task guides that are not in `CONFIG_MAPPING_NAMES` (therefore not in any
# `MODEL_MAPPING_NAMES` or any `MODEL_FOR_XXX_MAPPING_NAMES`).
__a :Optional[Any] = {
'summarization.md': ('nllb',),
'translation.md': ('nllb',),
}
def __snake_case ( __UpperCamelCase : Tuple ):
"""simple docstring"""
A_ = TASK_GUIDE_TO_MODELS[task_guide]
A_ = SPECIAL_TASK_GUIDE_TO_MODEL_TYPES.get(__UpperCamelCase ,set() )
A_ = {
code: name
for code, name in transformers_module.MODEL_NAMES_MAPPING.items()
if (code in model_maping_names or code in special_model_types)
}
return ", ".join([f'''[{name}](../model_doc/{code})''' for code, name in model_names.items()] ) + "\n"
def __snake_case ( __UpperCamelCase : str ,__UpperCamelCase : List[str]=False ):
"""simple docstring"""
A_ , A_ , A_ , A_ = _find_text_in_file(
filename=os.path.join(__UpperCamelCase ,__UpperCamelCase ) ,start_prompt="<!--This tip is automatically generated by `make fix-copies`, do not fill manually!-->" ,end_prompt="<!--End of the generated tip-->" ,)
A_ = get_model_list_for_task(__UpperCamelCase )
if current_list != new_list:
if overwrite:
with open(os.path.join(__UpperCamelCase ,__UpperCamelCase ) ,"w" ,encoding="utf-8" ,newline="\n" ) as f:
f.writelines(lines[:start_index] + [new_list] + lines[end_index:] )
else:
raise ValueError(
f'''The list of models that can be used in the {task_guide} guide needs an update. Run `make fix-copies`'''
" to fix this." )
if __name__ == "__main__":
__a :int = argparse.ArgumentParser()
parser.add_argument('--fix_and_overwrite', action='store_true', help='Whether to fix inconsistencies.')
__a :Optional[Any] = parser.parse_args()
for task_guide in TASK_GUIDE_TO_MODELS.keys():
check_model_list_for_task(task_guide, args.fix_and_overwrite) | 86 |
import logging
import os
from typing import List, Tuple
import numpy as np
import psutil
import torch
import torch.distributed as dist
from transformers import RagRetriever
__a :Any = logging.getLogger(__name__)
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Dict , UpperCAmelCase : List[Any] , UpperCAmelCase : Optional[int] , UpperCAmelCase : Optional[Any] , UpperCAmelCase : str=None ):
super().__init__(
UpperCAmelCase , question_encoder_tokenizer=UpperCAmelCase , generator_tokenizer=UpperCAmelCase , index=UpperCAmelCase , init_retrieval=UpperCAmelCase , )
A_ = None
def __A ( self : Dict , UpperCAmelCase : int ):
logger.info("initializing retrieval" )
# initializing a separate process group for retrieval as the default
# nccl backend doesn't support gather/scatter operations while gloo
# is too slow to replace nccl for the core gpu communication
if dist.is_initialized():
logger.info("dist initialized" )
# needs to be set manually
A_ = self._infer_socket_ifname()
# avoid clash with the NCCL port
A_ = str(distributed_port + 1 )
A_ = dist.new_group(ranks=UpperCAmelCase , backend="gloo" )
# initialize retriever only on the main worker
if not dist.is_initialized() or self._is_main():
logger.info("dist not initialized / main" )
self.index.init_index()
# all processes wait untill the retriever is initialized by the main process
if dist.is_initialized():
torch.distributed.barrier(group=self.process_group )
def __A ( self : List[str] ):
return dist.get_rank(group=self.process_group ) == 0
def __A ( self : Any , UpperCAmelCase : List[Any] , UpperCAmelCase : Union[str, Any] , UpperCAmelCase : Dict=torch.floataa ):
A_ = torch.empty(UpperCAmelCase , dtype=UpperCAmelCase )
dist.scatter(UpperCAmelCase , src=0 , scatter_list=UpperCAmelCase , group=self.process_group )
return target_tensor
def __A ( self : Any ):
A_ = psutil.net_if_addrs()
# a hacky way to deal with varying network interface names
A_ = next((addr for addr in addrs if addr.startswith("e" )) , UpperCAmelCase )
return ifname
def __A ( self : Tuple , UpperCAmelCase : np.ndarray , UpperCAmelCase : int ):
# single GPU training
if not dist.is_initialized():
A_ , A_ = self._main_retrieve(UpperCAmelCase , UpperCAmelCase )
return retrieved_doc_embeds, doc_ids, self.index.get_doc_dicts(UpperCAmelCase )
# distributed training
A_ = dist.get_world_size(group=self.process_group )
# gather logic
A_ = None
if self._is_main():
A_ = [torch.empty(question_hidden_states.shape , dtype=torch.floataa ) for _ in range(UpperCAmelCase )]
dist.gather(torch.tensor(UpperCAmelCase ) , dst=0 , gather_list=UpperCAmelCase , group=self.process_group )
# scatter logic
A_ = question_hidden_states.shape[0]
A_ = []
A_ = []
if self._is_main():
assert len(UpperCAmelCase ) == world_size
A_ , A_ = self._main_retrieve(torch.cat(UpperCAmelCase ).numpy() , UpperCAmelCase )
A_ , A_ = torch.tensor(UpperCAmelCase ), torch.tensor(UpperCAmelCase )
A_ = self._chunk_tensor(UpperCAmelCase , UpperCAmelCase )
A_ = self._chunk_tensor(UpperCAmelCase , UpperCAmelCase )
A_ = self._scattered(UpperCAmelCase , [n_queries, n_docs] , target_type=torch.intaa )
A_ = self._scattered(UpperCAmelCase , [n_queries, n_docs, question_hidden_states.shape[1]] )
return retrieved_doc_embeds.numpy(), doc_ids.numpy(), self.index.get_doc_dicts(UpperCAmelCase ) | 86 | 1 |
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : Dict ):
"""simple docstring"""
return (pointa[0] - pointa[0]) ** 2 + (pointa[1] - pointa[1]) ** 2
def __snake_case ( __UpperCamelCase : List[Any] ,__UpperCamelCase : List[Any]=0 ):
"""simple docstring"""
return sorted(__UpperCamelCase ,key=lambda __UpperCamelCase : x[column] )
def __snake_case ( __UpperCamelCase : Optional[int] ,__UpperCamelCase : str ,__UpperCamelCase : Optional[int]=float("inf" ) ):
"""simple docstring"""
for i in range(points_counts - 1 ):
for j in range(i + 1 ,__UpperCamelCase ):
A_ = euclidean_distance_sqr(points[i] ,points[j] )
if current_dis < min_dis:
A_ = current_dis
return min_dis
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : str ,__UpperCamelCase : Optional[int]=float("inf" ) ):
"""simple docstring"""
for i in range(min(6 ,points_counts - 1 ) ,__UpperCamelCase ):
for j in range(max(0 ,i - 6 ) ,__UpperCamelCase ):
A_ = euclidean_distance_sqr(points[i] ,points[j] )
if current_dis < min_dis:
A_ = current_dis
return min_dis
def __snake_case ( __UpperCamelCase : Union[str, Any] ,__UpperCamelCase : Optional[int] ,__UpperCamelCase : List[Any] ):
"""simple docstring"""
if points_counts <= 3:
return dis_between_closest_pair(__UpperCamelCase ,__UpperCamelCase )
# recursion
A_ = points_counts // 2
A_ = closest_pair_of_points_sqr(
__UpperCamelCase ,points_sorted_on_y[:mid] ,__UpperCamelCase )
A_ = closest_pair_of_points_sqr(
__UpperCamelCase ,points_sorted_on_y[mid:] ,points_counts - mid )
A_ = min(__UpperCamelCase ,__UpperCamelCase )
A_ = []
for point in points_sorted_on_x:
if abs(point[0] - points_sorted_on_x[mid][0] ) < closest_pair_dis:
cross_strip.append(__UpperCamelCase )
A_ = dis_between_closest_in_strip(
__UpperCamelCase ,len(__UpperCamelCase ) ,__UpperCamelCase )
return min(__UpperCamelCase ,__UpperCamelCase )
def __snake_case ( __UpperCamelCase : Union[str, Any] ,__UpperCamelCase : int ):
"""simple docstring"""
A_ = column_based_sort(__UpperCamelCase ,column=0 )
A_ = column_based_sort(__UpperCamelCase ,column=1 )
return (
closest_pair_of_points_sqr(
__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
) ** 0.5
if __name__ == "__main__":
__a :List[Any] = [(2, 3), (12, 30), (40, 50), (5, 1), (12, 10), (3, 4)]
print('Distance:', closest_pair_of_points(points, len(points))) | 86 |
from jiwer import compute_measures
import datasets
__a :List[Any] = '\\n@inproceedings{inproceedings,\n author = {Morris, Andrew and Maier, Viktoria and Green, Phil},\n year = {2004},\n month = {01},\n pages = {},\n title = {From WER and RIL to MER and WIL: improved evaluation measures for connected speech recognition.}\n}\n'
__a :Union[str, Any] = '\\nWord error rate (WER) is a common metric of the performance of an automatic speech recognition system.\n\nThe general difficulty of measuring performance lies in the fact that the recognized word sequence can have a different length from the reference word sequence (supposedly the correct one). The WER is derived from the Levenshtein distance, working at the word level instead of the phoneme level. The WER is a valuable tool for comparing different systems as well as for evaluating improvements within one system. This kind of measurement, however, provides no details on the nature of translation errors and further work is therefore required to identify the main source(s) of error and to focus any research effort.\n\nThis problem is solved by first aligning the recognized word sequence with the reference (spoken) word sequence using dynamic string alignment. Examination of this issue is seen through a theory called the power law that states the correlation between perplexity and word error rate.\n\nWord error rate can then be computed as:\n\nWER = (S + D + I) / N = (S + D + I) / (S + D + C)\n\nwhere\n\nS is the number of substitutions,\nD is the number of deletions,\nI is the number of insertions,\nC is the number of correct words,\nN is the number of words in the reference (N=S+D+C).\n\nThis value indicates the average number of errors per reference word. The lower the value, the better the\nperformance of the ASR system with a WER of 0 being a perfect score.\n'
__a :str = '\nCompute WER score of transcribed segments against references.\n\nArgs:\n references: List of references for each speech input.\n predictions: List of transcriptions to score.\n concatenate_texts (bool, default=False): Whether to concatenate all input texts or compute WER iteratively.\n\nReturns:\n (float): the word error rate\n\nExamples:\n\n >>> predictions = ["this is the prediction", "there is an other sample"]\n >>> references = ["this is the reference", "there is another one"]\n >>> wer = datasets.load_metric("wer")\n >>> wer_score = wer.compute(predictions=predictions, references=references)\n >>> print(wer_score)\n 0.5\n'
@datasets.utils.file_utils.add_start_docstrings(_DESCRIPTION , _KWARGS_DESCRIPTION )
class _a ( datasets.Metric ):
"""simple docstring"""
def __A ( self : Any ):
return datasets.MetricInfo(
description=_DESCRIPTION , citation=_CITATION , inputs_description=_KWARGS_DESCRIPTION , features=datasets.Features(
{
"predictions": datasets.Value("string" , id="sequence" ),
"references": datasets.Value("string" , id="sequence" ),
} ) , codebase_urls=["https://github.com/jitsi/jiwer/"] , reference_urls=[
"https://en.wikipedia.org/wiki/Word_error_rate",
] , )
def __A ( self : Dict , UpperCAmelCase : Dict=None , UpperCAmelCase : List[Any]=None , UpperCAmelCase : str=False ):
if concatenate_texts:
return compute_measures(UpperCAmelCase , UpperCAmelCase )["wer"]
else:
A_ = 0
A_ = 0
for prediction, reference in zip(UpperCAmelCase , UpperCAmelCase ):
A_ = compute_measures(UpperCAmelCase , UpperCAmelCase )
incorrect += measures["substitutions"] + measures["deletions"] + measures["insertions"]
total += measures["substitutions"] + measures["deletions"] + measures["hits"]
return incorrect / total | 86 | 1 |
from collections.abc import Callable
import numpy as np
def __snake_case ( __UpperCamelCase : Callable ,__UpperCamelCase : float ,__UpperCamelCase : float ,__UpperCamelCase : float ,__UpperCamelCase : float ):
"""simple docstring"""
A_ = int(np.ceil((x_end - xa) / step_size ) )
A_ = np.zeros((n + 1,) )
A_ = ya
A_ = xa
for k in range(__UpperCamelCase ):
A_ = y[k] + step_size * ode_func(__UpperCamelCase ,y[k] )
A_ = y[k] + (
(step_size / 2) * (ode_func(__UpperCamelCase ,y[k] ) + ode_func(x + step_size ,__UpperCamelCase ))
)
x += step_size
return y
if __name__ == "__main__":
import doctest
doctest.testmod() | 86 |
class _a :
"""simple docstring"""
def __init__( self : Union[str, Any] , UpperCAmelCase : int , UpperCAmelCase : Any , UpperCAmelCase : Dict ):
A_ = None
A_ = None
A_ = graph
self._normalize_graph(UpperCAmelCase , UpperCAmelCase )
A_ = len(UpperCAmelCase )
A_ = None
def __A ( self : Union[str, Any] , UpperCAmelCase : List[str] , UpperCAmelCase : Tuple ):
if sources is int:
A_ = [sources]
if sinks is int:
A_ = [sinks]
if len(UpperCAmelCase ) == 0 or len(UpperCAmelCase ) == 0:
return
A_ = sources[0]
A_ = sinks[0]
# make fake vertex if there are more
# than one source or sink
if len(UpperCAmelCase ) > 1 or len(UpperCAmelCase ) > 1:
A_ = 0
for i in sources:
max_input_flow += sum(self.graph[i] )
A_ = len(self.graph ) + 1
for room in self.graph:
room.insert(0 , 0 )
self.graph.insert(0 , [0] * size )
for i in sources:
A_ = max_input_flow
A_ = 0
A_ = len(self.graph ) + 1
for room in self.graph:
room.append(0 )
self.graph.append([0] * size )
for i in sinks:
A_ = max_input_flow
A_ = size - 1
def __A ( self : str ):
if self.maximum_flow_algorithm is None:
raise Exception("You need to set maximum flow algorithm before." )
if self.source_index is None or self.sink_index is None:
return 0
self.maximum_flow_algorithm.execute()
return self.maximum_flow_algorithm.getMaximumFlow()
def __A ( self : Tuple , UpperCAmelCase : List[Any] ):
A_ = algorithm(self )
class _a :
"""simple docstring"""
def __init__( self : Tuple , UpperCAmelCase : List[str] ):
A_ = flow_network
A_ = flow_network.verticesCount
A_ = flow_network.sourceIndex
A_ = flow_network.sinkIndex
# it's just a reference, so you shouldn't change
# it in your algorithms, use deep copy before doing that
A_ = flow_network.graph
A_ = False
def __A ( self : Optional[int] ):
if not self.executed:
self._algorithm()
A_ = True
def __A ( self : Dict ):
pass
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Optional[Any] , UpperCAmelCase : List[Any] ):
super().__init__(UpperCAmelCase )
# use this to save your result
A_ = -1
def __A ( self : Tuple ):
if not self.executed:
raise Exception("You should execute algorithm before using its result!" )
return self.maximum_flow
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Tuple , UpperCAmelCase : Union[str, Any] ):
super().__init__(UpperCAmelCase )
A_ = [[0] * self.verticies_count for i in range(self.verticies_count )]
A_ = [0] * self.verticies_count
A_ = [0] * self.verticies_count
def __A ( self : List[str] ):
A_ = self.verticies_count
# push some substance to graph
for nextvertex_index, bandwidth in enumerate(self.graph[self.source_index] ):
self.preflow[self.source_index][nextvertex_index] += bandwidth
self.preflow[nextvertex_index][self.source_index] -= bandwidth
self.excesses[nextvertex_index] += bandwidth
# Relabel-to-front selection rule
A_ = [
i
for i in range(self.verticies_count )
if i != self.source_index and i != self.sink_index
]
# move through list
A_ = 0
while i < len(UpperCAmelCase ):
A_ = vertices_list[i]
A_ = self.heights[vertex_index]
self.process_vertex(UpperCAmelCase )
if self.heights[vertex_index] > previous_height:
# if it was relabeled, swap elements
# and start from 0 index
vertices_list.insert(0 , vertices_list.pop(UpperCAmelCase ) )
A_ = 0
else:
i += 1
A_ = sum(self.preflow[self.source_index] )
def __A ( self : List[str] , UpperCAmelCase : Dict ):
while self.excesses[vertex_index] > 0:
for neighbour_index in range(self.verticies_count ):
# if it's neighbour and current vertex is higher
if (
self.graph[vertex_index][neighbour_index]
- self.preflow[vertex_index][neighbour_index]
> 0
and self.heights[vertex_index] > self.heights[neighbour_index]
):
self.push(UpperCAmelCase , UpperCAmelCase )
self.relabel(UpperCAmelCase )
def __A ( self : Optional[Any] , UpperCAmelCase : Dict , UpperCAmelCase : str ):
A_ = min(
self.excesses[from_index] , self.graph[from_index][to_index] - self.preflow[from_index][to_index] , )
self.preflow[from_index][to_index] += preflow_delta
self.preflow[to_index][from_index] -= preflow_delta
self.excesses[from_index] -= preflow_delta
self.excesses[to_index] += preflow_delta
def __A ( self : Optional[Any] , UpperCAmelCase : List[Any] ):
A_ = None
for to_index in range(self.verticies_count ):
if (
self.graph[vertex_index][to_index]
- self.preflow[vertex_index][to_index]
> 0
) and (min_height is None or self.heights[to_index] < min_height):
A_ = self.heights[to_index]
if min_height is not None:
A_ = min_height + 1
if __name__ == "__main__":
__a :Tuple = [0]
__a :Tuple = [3]
# graph = [
# [0, 0, 4, 6, 0, 0],
# [0, 0, 5, 2, 0, 0],
# [0, 0, 0, 0, 4, 4],
# [0, 0, 0, 0, 6, 6],
# [0, 0, 0, 0, 0, 0],
# [0, 0, 0, 0, 0, 0],
# ]
__a :List[str] = [[0, 7, 0, 0], [0, 0, 6, 0], [0, 0, 0, 8], [9, 0, 0, 0]]
# prepare our network
__a :List[str] = FlowNetwork(graph, entrances, exits)
# set algorithm
flow_network.set_maximum_flow_algorithm(PushRelabelExecutor)
# and calculate
__a :List[Any] = flow_network.find_maximum_flow()
print(F"maximum flow is {maximum_flow}") | 86 | 1 |
from ..utils import DummyObject, requires_backends
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : List[Any] = ['torch', 'torchsde']
def __init__( self : Any , *UpperCAmelCase : int , **UpperCAmelCase : int ):
requires_backends(self , ["torch", "torchsde"] )
@classmethod
def __A ( cls : Optional[int] , *UpperCAmelCase : Any , **UpperCAmelCase : int ):
requires_backends(cls , ["torch", "torchsde"] )
@classmethod
def __A ( cls : int , *UpperCAmelCase : Tuple , **UpperCAmelCase : List[Any] ):
requires_backends(cls , ["torch", "torchsde"] ) | 86 |
from typing import TYPE_CHECKING
from ...utils import (
OptionalDependencyNotAvailable,
_LazyModule,
is_flax_available,
is_sentencepiece_available,
is_tf_available,
is_tokenizers_available,
is_torch_available,
)
__a :Dict = {'configuration_xglm': ['XGLM_PRETRAINED_CONFIG_ARCHIVE_MAP', 'XGLMConfig']}
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Dict = ['XGLMTokenizer']
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :str = ['XGLMTokenizerFast']
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Tuple = [
'XGLM_PRETRAINED_MODEL_ARCHIVE_LIST',
'XGLMForCausalLM',
'XGLMModel',
'XGLMPreTrainedModel',
]
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :List[Any] = [
'FlaxXGLMForCausalLM',
'FlaxXGLMModel',
'FlaxXGLMPreTrainedModel',
]
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Any = [
'TF_XGLM_PRETRAINED_MODEL_ARCHIVE_LIST',
'TFXGLMForCausalLM',
'TFXGLMModel',
'TFXGLMPreTrainedModel',
]
if TYPE_CHECKING:
from .configuration_xglm import XGLM_PRETRAINED_CONFIG_ARCHIVE_MAP, XGLMConfig
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_xglm import XGLMTokenizer
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_xglm_fast import XGLMTokenizerFast
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_xglm import XGLM_PRETRAINED_MODEL_ARCHIVE_LIST, XGLMForCausalLM, XGLMModel, XGLMPreTrainedModel
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_flax_xglm import FlaxXGLMForCausalLM, FlaxXGLMModel, FlaxXGLMPreTrainedModel
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tf_xglm import (
TF_XGLM_PRETRAINED_MODEL_ARCHIVE_LIST,
TFXGLMForCausalLM,
TFXGLMModel,
TFXGLMPreTrainedModel,
)
else:
import sys
__a :List[Any] = _LazyModule(__name__, globals()['__file__'], _import_structure) | 86 | 1 |
from __future__ import annotations
from collections.abc import Callable
from typing import Generic, TypeVar
SCREAMING_SNAKE_CASE__ : str = TypeVar("""T""")
SCREAMING_SNAKE_CASE__ : int = TypeVar("""U""")
class lowerCamelCase_ ( Generic[T, U] ):
def __init__( self , __lowerCAmelCase , __lowerCAmelCase ):
"""simple docstring"""
__magic_name__ :Tuple = key
__magic_name__ :List[str] = val
__magic_name__ :DoubleLinkedListNode[T, U] | None = None
__magic_name__ :DoubleLinkedListNode[T, U] | None = None
def __repr__( self ):
"""simple docstring"""
return (
F'''Node: key: {self.key}, val: {self.val}, '''
F'''has next: {bool(self.next )}, has prev: {bool(self.prev )}'''
)
class lowerCamelCase_ ( Generic[T, U] ):
def __init__( self ):
"""simple docstring"""
__magic_name__ :DoubleLinkedListNode[T, U] = DoubleLinkedListNode(__lowerCAmelCase , __lowerCAmelCase )
__magic_name__ :DoubleLinkedListNode[T, U] = DoubleLinkedListNode(__lowerCAmelCase , __lowerCAmelCase )
__magic_name__ , __magic_name__ :Union[str, Any] = self.rear, self.head
def __repr__( self ):
"""simple docstring"""
__magic_name__ :Any = ['''DoubleLinkedList''']
__magic_name__ :Any = self.head
while node.next is not None:
rep.append(str(__lowerCAmelCase ) )
__magic_name__ :Optional[int] = node.next
rep.append(str(self.rear ) )
return ",\n ".join(__lowerCAmelCase )
def A ( self , __lowerCAmelCase ):
"""simple docstring"""
__magic_name__ :List[Any] = self.rear.prev
# All nodes other than self.head are guaranteed to have non-None previous
assert previous is not None
__magic_name__ :str = node
__magic_name__ :str = previous
__magic_name__ :Dict = node
__magic_name__ :Optional[Any] = self.rear
def A ( self , __lowerCAmelCase ):
"""simple docstring"""
if node.prev is None or node.next is None:
return None
__magic_name__ :str = node.next
__magic_name__ :Any = node.prev
__magic_name__ :int = None
__magic_name__ :List[str] = None
return node
class lowerCamelCase_ ( Generic[T, U] ):
a__ = {}
def __init__( self , __lowerCAmelCase ):
"""simple docstring"""
__magic_name__ :DoubleLinkedList[T, U] = DoubleLinkedList()
__magic_name__ :Dict = capacity
__magic_name__ :Union[str, Any] = 0
__magic_name__ :Dict = 0
__magic_name__ :Optional[Any] = 0
__magic_name__ :dict[T, DoubleLinkedListNode[T, U]] = {}
def __repr__( self ):
"""simple docstring"""
return (
F'''CacheInfo(hits={self.hits}, misses={self.miss}, '''
F'''capacity={self.capacity}, current size={self.num_keys})'''
)
def __contains__( self , __lowerCAmelCase ):
"""simple docstring"""
return key in self.cache
def A ( self , __lowerCAmelCase ):
"""simple docstring"""
# Note: pythonic interface would throw KeyError rather than return None
if key in self.cache:
self.hits += 1
__magic_name__ :DoubleLinkedListNode[T, U] = self.cache[key]
__magic_name__ :int = self.list.remove(self.cache[key] )
assert node == value_node
# node is guaranteed not None because it is in self.cache
assert node is not None
self.list.add(__lowerCAmelCase )
return node.val
self.miss += 1
return None
def A ( self , __lowerCAmelCase , __lowerCAmelCase ):
"""simple docstring"""
if key not in self.cache:
if self.num_keys >= self.capacity:
# delete first node (oldest) when over capacity
__magic_name__ :Union[str, Any] = self.list.head.next
# guaranteed to have a non-None first node when num_keys > 0
# explain to type checker via assertions
assert first_node is not None
assert first_node.key is not None
assert (
self.list.remove(__lowerCAmelCase ) is not None
) # node guaranteed to be in list assert node.key is not None
del self.cache[first_node.key]
self.num_keys -= 1
__magic_name__ :List[Any] = DoubleLinkedListNode(__lowerCAmelCase , __lowerCAmelCase )
self.list.add(self.cache[key] )
self.num_keys += 1
else:
# bump node to the end of the list, update value
__magic_name__ :str = self.list.remove(self.cache[key] )
assert node is not None # node guaranteed to be in list
__magic_name__ :Any = value
self.list.add(__lowerCAmelCase )
@classmethod
def A ( cls , __lowerCAmelCase = 1_2_8 ):
"""simple docstring"""
def cache_decorator_inner(__lowerCAmelCase ) -> Callable[..., U]:
def cache_decorator_wrapper(*__lowerCAmelCase ) -> U:
if func not in cls.decorator_function_to_instance_map:
__magic_name__ :List[str] = LRUCache(__lowerCAmelCase )
__magic_name__ :Optional[Any] = cls.decorator_function_to_instance_map[func].get(args[0] )
if result is None:
__magic_name__ :Optional[Any] = func(*__lowerCAmelCase )
cls.decorator_function_to_instance_map[func].put(args[0] , __lowerCAmelCase )
return result
def cache_info() -> LRUCache[T, U]:
return cls.decorator_function_to_instance_map[func]
setattr(__lowerCAmelCase , '''cache_info''' , __lowerCAmelCase ) # noqa: B010
return cache_decorator_wrapper
return cache_decorator_inner
if __name__ == "__main__":
import doctest
doctest.testmod()
| 0 |
# Usage:
# ./gen-card-facebook-wmt19.py
import os
from pathlib import Path
def __snake_case ( __UpperCamelCase : str ,__UpperCamelCase : str ,__UpperCamelCase : List[str] ):
"""simple docstring"""
A_ = {
"en": "Machine learning is great, isn't it?",
"ru": "Машинное обучение - это здорово, не так ли?",
"de": "Maschinelles Lernen ist großartig, oder?",
}
# BLUE scores as follows:
# "pair": [fairseq, transformers]
A_ = {
"ru-en": ["[41.3](http://matrix.statmt.org/matrix/output/1907?run_id=6937)", "39.20"],
"en-ru": ["[36.4](http://matrix.statmt.org/matrix/output/1914?run_id=6724)", "33.47"],
"en-de": ["[43.1](http://matrix.statmt.org/matrix/output/1909?run_id=6862)", "42.83"],
"de-en": ["[42.3](http://matrix.statmt.org/matrix/output/1902?run_id=6750)", "41.35"],
}
A_ = f'''{src_lang}-{tgt_lang}'''
A_ = f'''
---
language:
- {src_lang}
- {tgt_lang}
thumbnail:
tags:
- translation
- wmt19
- facebook
license: apache-2.0
datasets:
- wmt19
metrics:
- bleu
---
# FSMT
## Model description
This is a ported version of [fairseq wmt19 transformer](https://github.com/pytorch/fairseq/blob/master/examples/wmt19/README.md) for {src_lang}-{tgt_lang}.
For more details, please see, [Facebook FAIR\'s WMT19 News Translation Task Submission](https://arxiv.org/abs/1907.06616).
The abbreviation FSMT stands for FairSeqMachineTranslation
All four models are available:
* [wmt19-en-ru](https://huggingface.co/facebook/wmt19-en-ru)
* [wmt19-ru-en](https://huggingface.co/facebook/wmt19-ru-en)
* [wmt19-en-de](https://huggingface.co/facebook/wmt19-en-de)
* [wmt19-de-en](https://huggingface.co/facebook/wmt19-de-en)
## Intended uses & limitations
#### How to use
```python
from transformers import FSMTForConditionalGeneration, FSMTTokenizer
mname = "facebook/wmt19-{src_lang}-{tgt_lang}"
tokenizer = FSMTTokenizer.from_pretrained(mname)
model = FSMTForConditionalGeneration.from_pretrained(mname)
input = "{texts[src_lang]}"
input_ids = tokenizer.encode(input, return_tensors="pt")
outputs = model.generate(input_ids)
decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(decoded) # {texts[tgt_lang]}
```
#### Limitations and bias
- The original (and this ported model) doesn\'t seem to handle well inputs with repeated sub-phrases, [content gets truncated](https://discuss.huggingface.co/t/issues-with-translating-inputs-containing-repeated-phrases/981)
## Training data
Pretrained weights were left identical to the original model released by fairseq. For more details, please, see the [paper](https://arxiv.org/abs/1907.06616).
## Eval results
pair | fairseq | transformers
-------|---------|----------
{pair} | {scores[pair][0]} | {scores[pair][1]}
The score is slightly below the score reported by `fairseq`, since `transformers`` currently doesn\'t support:
- model ensemble, therefore the best performing checkpoint was ported (``model4.pt``).
- re-ranking
The score was calculated using this code:
```bash
git clone https://github.com/huggingface/transformers
cd transformers
export PAIR={pair}
export DATA_DIR=data/$PAIR
export SAVE_DIR=data/$PAIR
export BS=8
export NUM_BEAMS=15
mkdir -p $DATA_DIR
sacrebleu -t wmt19 -l $PAIR --echo src > $DATA_DIR/val.source
sacrebleu -t wmt19 -l $PAIR --echo ref > $DATA_DIR/val.target
echo $PAIR
PYTHONPATH="src:examples/seq2seq" python examples/seq2seq/run_eval.py facebook/wmt19-$PAIR $DATA_DIR/val.source $SAVE_DIR/test_translations.txt --reference_path $DATA_DIR/val.target --score_path $SAVE_DIR/test_bleu.json --bs $BS --task translation --num_beams $NUM_BEAMS
```
note: fairseq reports using a beam of 50, so you should get a slightly higher score if re-run with `--num_beams 50`.
## Data Sources
- [training, etc.](http://www.statmt.org/wmt19/)
- [test set](http://matrix.statmt.org/test_sets/newstest2019.tgz?1556572561)
### BibTeX entry and citation info
```bibtex
@inproceedings{{...,
year={{2020}},
title={{Facebook FAIR\'s WMT19 News Translation Task Submission}},
author={{Ng, Nathan and Yee, Kyra and Baevski, Alexei and Ott, Myle and Auli, Michael and Edunov, Sergey}},
booktitle={{Proc. of WMT}},
}}
```
## TODO
- port model ensemble (fairseq uses 4 model checkpoints)
'''
os.makedirs(__UpperCamelCase ,exist_ok=__UpperCamelCase )
A_ = os.path.join(__UpperCamelCase ,"README.md" )
print(f'''Generating {path}''' )
with open(__UpperCamelCase ,"w" ,encoding="utf-8" ) as f:
f.write(__UpperCamelCase )
# make sure we are under the root of the project
__a :Optional[Any] = Path(__file__).resolve().parent.parent.parent
__a :Optional[Any] = repo_dir / 'model_cards'
for model_name in ["wmt19-ru-en", "wmt19-en-ru", "wmt19-en-de", "wmt19-de-en"]:
__a , __a , __a :int = model_name.split('-')
__a :str = model_cards_dir / 'facebook' / model_name
write_model_card(model_card_dir, src_lang=src_lang, tgt_lang=tgt_lang) | 86 | 0 |
import torch
from diffusers import DPMSolverSDEScheduler
from diffusers.utils import torch_device
from diffusers.utils.testing_utils import require_torchsde
from .test_schedulers import SchedulerCommonTest
@require_torchsde
class __lowerCamelCase (_a ):
_lowercase = (DPMSolverSDEScheduler,)
_lowercase = 10
def snake_case_ ( self: Tuple,**A_: List[str] ):
'''simple docstring'''
__UpperCamelCase = {
'num_train_timesteps': 1100,
'beta_start': 0.0_0_0_1,
'beta_end': 0.0_2,
'beta_schedule': 'linear',
'noise_sampler_seed': 0,
}
config.update(**A_ )
return config
def snake_case_ ( self: List[Any] ):
'''simple docstring'''
for timesteps in [10, 50, 100, 1000]:
self.check_over_configs(num_train_timesteps=A_ )
def snake_case_ ( self: str ):
'''simple docstring'''
for beta_start, beta_end in zip([0.0_0_0_0_1, 0.0_0_0_1, 0.0_0_1],[0.0_0_0_2, 0.0_0_2, 0.0_2] ):
self.check_over_configs(beta_start=A_,beta_end=A_ )
def snake_case_ ( self: Dict ):
'''simple docstring'''
for schedule in ["linear", "scaled_linear"]:
self.check_over_configs(beta_schedule=A_ )
def snake_case_ ( self: List[str] ):
'''simple docstring'''
for prediction_type in ["epsilon", "v_prediction"]:
self.check_over_configs(prediction_type=A_ )
def snake_case_ ( self: Union[str, Any] ):
'''simple docstring'''
__UpperCamelCase = self.scheduler_classes[0]
__UpperCamelCase = self.get_scheduler_config()
__UpperCamelCase = scheduler_class(**A_ )
scheduler.set_timesteps(self.num_inference_steps )
__UpperCamelCase = self.dummy_model()
__UpperCamelCase = self.dummy_sample_deter * scheduler.init_noise_sigma
__UpperCamelCase = sample.to(A_ )
for i, t in enumerate(scheduler.timesteps ):
__UpperCamelCase = scheduler.scale_model_input(A_,A_ )
__UpperCamelCase = model(A_,A_ )
__UpperCamelCase = scheduler.step(A_,A_,A_ )
__UpperCamelCase = output.prev_sample
__UpperCamelCase = torch.sum(torch.abs(A_ ) )
__UpperCamelCase = torch.mean(torch.abs(A_ ) )
if torch_device in ["mps"]:
assert abs(result_sum.item() - 1_6_7.4_7_8_2_1_0_4_4_9_2_1_8_7_5 ) < 1E-2
assert abs(result_mean.item() - 0.2_1_7_8_7_0_5_9_6_4_5_6_5_2_7_7 ) < 1E-3
elif torch_device in ["cuda"]:
assert abs(result_sum.item() - 1_7_1.5_9_3_5_2_1_1_1_8_1_6_4_0_6 ) < 1E-2
assert abs(result_mean.item() - 0.2_2_3_4_2_9_0_6_8_9_2_2_9_9_6_5_2 ) < 1E-3
else:
assert abs(result_sum.item() - 1_6_2.5_2_3_8_3_4_2_2_8_5_1_5_6_2 ) < 1E-2
assert abs(result_mean.item() - 0.2_1_1_6_1_9_5_7_0_8_5_1_3_2_6 ) < 1E-3
def snake_case_ ( self: str ):
'''simple docstring'''
__UpperCamelCase = self.scheduler_classes[0]
__UpperCamelCase = self.get_scheduler_config(prediction_type='v_prediction' )
__UpperCamelCase = scheduler_class(**A_ )
scheduler.set_timesteps(self.num_inference_steps )
__UpperCamelCase = self.dummy_model()
__UpperCamelCase = self.dummy_sample_deter * scheduler.init_noise_sigma
__UpperCamelCase = sample.to(A_ )
for i, t in enumerate(scheduler.timesteps ):
__UpperCamelCase = scheduler.scale_model_input(A_,A_ )
__UpperCamelCase = model(A_,A_ )
__UpperCamelCase = scheduler.step(A_,A_,A_ )
__UpperCamelCase = output.prev_sample
__UpperCamelCase = torch.sum(torch.abs(A_ ) )
__UpperCamelCase = torch.mean(torch.abs(A_ ) )
if torch_device in ["mps"]:
assert abs(result_sum.item() - 1_2_4.7_7_1_4_9_2_0_0_4_3_9_4_5_3 ) < 1E-2
assert abs(result_mean.item() - 0.1_6_2_2_6_2_8_9_0_1_4_8_1_6_2_8_4 ) < 1E-3
elif torch_device in ["cuda"]:
assert abs(result_sum.item() - 1_2_8.1_6_6_3_3_6_0_5_9_5_7_0_3 ) < 1E-2
assert abs(result_mean.item() - 0.1_6_6_8_8_3_2_6_0_0_1_1_6_7_2_9_7 ) < 1E-3
else:
assert abs(result_sum.item() - 1_1_9.8_4_8_7_5_4_8_8_2_8_1_2_5 ) < 1E-2
assert abs(result_mean.item() - 0.1_5_6_0_5_3_0_6_6_2_5_3_6_6_2_1 ) < 1E-3
def snake_case_ ( self: List[Any] ):
'''simple docstring'''
__UpperCamelCase = self.scheduler_classes[0]
__UpperCamelCase = self.get_scheduler_config()
__UpperCamelCase = scheduler_class(**A_ )
scheduler.set_timesteps(self.num_inference_steps,device=A_ )
__UpperCamelCase = self.dummy_model()
__UpperCamelCase = self.dummy_sample_deter.to(A_ ) * scheduler.init_noise_sigma
for t in scheduler.timesteps:
__UpperCamelCase = scheduler.scale_model_input(A_,A_ )
__UpperCamelCase = model(A_,A_ )
__UpperCamelCase = scheduler.step(A_,A_,A_ )
__UpperCamelCase = output.prev_sample
__UpperCamelCase = torch.sum(torch.abs(A_ ) )
__UpperCamelCase = torch.mean(torch.abs(A_ ) )
if torch_device in ["mps"]:
assert abs(result_sum.item() - 1_6_7.4_6_9_5_7_3_9_7_4_6_0_9_3_8 ) < 1E-2
assert abs(result_mean.item() - 0.2_1_8_0_5_9_3_4_6_0_7_9_8_2_6_3_5 ) < 1E-3
elif torch_device in ["cuda"]:
assert abs(result_sum.item() - 1_7_1.5_9_3_5_3_6_3_7_6_9_5_3_1_2 ) < 1E-2
assert abs(result_mean.item() - 0.2_2_3_4_2_9_0_8_3_8_2_4_1_5_7_7_1 ) < 1E-3
else:
assert abs(result_sum.item() - 1_6_2.5_2_3_8_3_4_2_2_8_5_1_5_6_2 ) < 1E-2
assert abs(result_mean.item() - 0.2_1_1_6_1_9_5_7_0_8_5_1_3_2_6 ) < 1E-3
def snake_case_ ( self: Dict ):
'''simple docstring'''
__UpperCamelCase = self.scheduler_classes[0]
__UpperCamelCase = self.get_scheduler_config()
__UpperCamelCase = scheduler_class(**A_,use_karras_sigmas=A_ )
scheduler.set_timesteps(self.num_inference_steps,device=A_ )
__UpperCamelCase = self.dummy_model()
__UpperCamelCase = self.dummy_sample_deter.to(A_ ) * scheduler.init_noise_sigma
__UpperCamelCase = sample.to(A_ )
for t in scheduler.timesteps:
__UpperCamelCase = scheduler.scale_model_input(A_,A_ )
__UpperCamelCase = model(A_,A_ )
__UpperCamelCase = scheduler.step(A_,A_,A_ )
__UpperCamelCase = output.prev_sample
__UpperCamelCase = torch.sum(torch.abs(A_ ) )
__UpperCamelCase = torch.mean(torch.abs(A_ ) )
if torch_device in ["mps"]:
assert abs(result_sum.item() - 1_7_6.6_6_9_7_4_1_3_5_7_4_2_1_8_8 ) < 1E-2
assert abs(result_mean.item() - 0.2_3_0_0_3_8_7_2_7_3_0_9_8_1_8_1_1 ) < 1E-2
elif torch_device in ["cuda"]:
assert abs(result_sum.item() - 1_7_7.6_3_6_5_3_5_6_4_4_5_3_1_2_5 ) < 1E-2
assert abs(result_mean.item() - 0.2_3_0_0_3_8_7_2_7_3_0_9_8_1_8_1_1 ) < 1E-2
else:
assert abs(result_sum.item() - 1_7_0.3_1_3_5_2_2_3_3_8_8_6_7_2 ) < 1E-2
assert abs(result_mean.item() - 0.2_3_0_0_3_8_7_2_7_3_0_9_8_1_8_1_1 ) < 1E-2
| 1 |
from ..utils import DummyObject, requires_backends
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Optional[Any] = ['torch', 'transformers', 'onnx']
def __init__( self : str , *UpperCAmelCase : int , **UpperCAmelCase : List[Any] ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Dict , *UpperCAmelCase : Dict , **UpperCAmelCase : Union[str, Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[Any] , *UpperCAmelCase : List[Any] , **UpperCAmelCase : Union[str, Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : str = ['torch', 'transformers', 'onnx']
def __init__( self : Optional[int] , *UpperCAmelCase : List[str] , **UpperCAmelCase : int ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : List[str] , *UpperCAmelCase : List[str] , **UpperCAmelCase : List[Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[Any] , *UpperCAmelCase : Dict , **UpperCAmelCase : Optional[Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Optional[Any] = ['torch', 'transformers', 'onnx']
def __init__( self : Union[str, Any] , *UpperCAmelCase : Any , **UpperCAmelCase : Dict ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Tuple , *UpperCAmelCase : Dict , **UpperCAmelCase : str ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Dict , *UpperCAmelCase : List[str] , **UpperCAmelCase : List[str] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : int = ['torch', 'transformers', 'onnx']
def __init__( self : List[str] , *UpperCAmelCase : Dict , **UpperCAmelCase : Optional[Any] ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Any , *UpperCAmelCase : Union[str, Any] , **UpperCAmelCase : Dict ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[Any] , *UpperCAmelCase : int , **UpperCAmelCase : Optional[int] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Dict = ['torch', 'transformers', 'onnx']
def __init__( self : List[str] , *UpperCAmelCase : str , **UpperCAmelCase : int ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : List[str] , *UpperCAmelCase : Optional[int] , **UpperCAmelCase : Optional[int] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[int] , *UpperCAmelCase : int , **UpperCAmelCase : List[Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : int = ['torch', 'transformers', 'onnx']
def __init__( self : Tuple , *UpperCAmelCase : List[Any] , **UpperCAmelCase : Optional[Any] ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[int] , *UpperCAmelCase : Dict , **UpperCAmelCase : str ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[int] , *UpperCAmelCase : Optional[Any] , **UpperCAmelCase : int ):
requires_backends(cls , ["torch", "transformers", "onnx"] ) | 86 | 0 |
from __future__ import annotations
import copy
import inspect
import unittest
import numpy as np
from transformers import is_tf_available, is_vision_available
from transformers.models.auto import get_values
from transformers.testing_utils import require_tf, slow
from transformers.utils import cached_property
from ...test_configuration_common import ConfigTester
from ...test_modeling_tf_common import TFModelTesterMixin, floats_tensor, ids_tensor, random_attention_mask
from ...test_pipeline_mixin import PipelineTesterMixin
if is_tf_available():
import tensorflow as tf
from transformers import (
TF_LAYOUTLMV3_PRETRAINED_MODEL_ARCHIVE_LIST,
TF_MODEL_FOR_MULTIPLE_CHOICE_MAPPING,
TF_MODEL_FOR_QUESTION_ANSWERING_MAPPING,
TF_MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING,
TF_MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING,
LayoutLMvaConfig,
TFLayoutLMvaForQuestionAnswering,
TFLayoutLMvaForSequenceClassification,
TFLayoutLMvaForTokenClassification,
TFLayoutLMvaModel,
)
if is_vision_available():
from PIL import Image
from transformers import LayoutLMvaImageProcessor
class lowerCamelCase__ :
"""simple docstring"""
def __init__( self : str , __lowerCAmelCase : List[Any] , __lowerCAmelCase : Union[str, Any]=2 , __lowerCAmelCase : int=3 , __lowerCAmelCase : Tuple=4 , __lowerCAmelCase : Union[str, Any]=2 , __lowerCAmelCase : List[str]=7 , __lowerCAmelCase : List[Any]=True , __lowerCAmelCase : str=True , __lowerCAmelCase : Optional[Any]=True , __lowerCAmelCase : Any=True , __lowerCAmelCase : str=99 , __lowerCAmelCase : int=36 , __lowerCAmelCase : List[Any]=2 , __lowerCAmelCase : List[Any]=4 , __lowerCAmelCase : Dict=37 , __lowerCAmelCase : Dict="gelu" , __lowerCAmelCase : Optional[int]=0.1 , __lowerCAmelCase : Tuple=0.1 , __lowerCAmelCase : Optional[Any]=5_12 , __lowerCAmelCase : Any=16 , __lowerCAmelCase : Optional[Any]=2 , __lowerCAmelCase : Optional[int]=0.02 , __lowerCAmelCase : int=6 , __lowerCAmelCase : Tuple=6 , __lowerCAmelCase : Dict=3 , __lowerCAmelCase : Dict=4 , __lowerCAmelCase : Optional[Any]=None , __lowerCAmelCase : Tuple=10_00 , ) -> List[str]:
_A = parent
_A = batch_size
_A = num_channels
_A = image_size
_A = patch_size
_A = is_training
_A = use_input_mask
_A = use_token_type_ids
_A = use_labels
_A = vocab_size
_A = hidden_size
_A = num_hidden_layers
_A = num_attention_heads
_A = intermediate_size
_A = hidden_act
_A = hidden_dropout_prob
_A = attention_probs_dropout_prob
_A = max_position_embeddings
_A = type_vocab_size
_A = type_sequence_label_size
_A = initializer_range
_A = coordinate_size
_A = shape_size
_A = num_labels
_A = num_choices
_A = scope
_A = range_bbox
# LayoutLMv3's sequence length equals the number of text tokens + number of patches + 1 (we add 1 for the CLS token)
_A = text_seq_length
_A = (image_size // patch_size) ** 2 + 1
_A = self.text_seq_length + self.image_seq_length
def snake_case_ ( self : List[Any] ) -> Dict:
_A = ids_tensor([self.batch_size, self.text_seq_length] , self.vocab_size )
_A = ids_tensor([self.batch_size, self.text_seq_length, 4] , self.range_bbox )
_A = bbox.numpy()
# Ensure that bbox is legal
for i in range(bbox.shape[0] ):
for j in range(bbox.shape[1] ):
if bbox[i, j, 3] < bbox[i, j, 1]:
_A = bbox[i, j, 3]
_A = bbox[i, j, 1]
_A = tmp_coordinate
if bbox[i, j, 2] < bbox[i, j, 0]:
_A = bbox[i, j, 2]
_A = bbox[i, j, 0]
_A = tmp_coordinate
_A = tf.constant(__lowerCAmelCase )
_A = floats_tensor([self.batch_size, self.num_channels, self.image_size, self.image_size] )
_A = None
if self.use_input_mask:
_A = random_attention_mask([self.batch_size, self.text_seq_length] )
_A = None
if self.use_token_type_ids:
_A = ids_tensor([self.batch_size, self.text_seq_length] , self.type_vocab_size )
_A = None
_A = None
if self.use_labels:
_A = ids_tensor([self.batch_size] , self.type_sequence_label_size )
_A = ids_tensor([self.batch_size, self.text_seq_length] , self.num_labels )
_A = LayoutLMvaConfig(
vocab_size=self.vocab_size , hidden_size=self.hidden_size , num_hidden_layers=self.num_hidden_layers , num_attention_heads=self.num_attention_heads , intermediate_size=self.intermediate_size , hidden_act=self.hidden_act , hidden_dropout_prob=self.hidden_dropout_prob , attention_probs_dropout_prob=self.attention_probs_dropout_prob , max_position_embeddings=self.max_position_embeddings , type_vocab_size=self.type_vocab_size , initializer_range=self.initializer_range , coordinate_size=self.coordinate_size , shape_size=self.shape_size , input_size=self.image_size , patch_size=self.patch_size , )
return config, input_ids, bbox, pixel_values, token_type_ids, input_mask, sequence_labels, token_labels
def snake_case_ ( self : Union[str, Any] , __lowerCAmelCase : Dict , __lowerCAmelCase : Tuple , __lowerCAmelCase : List[Any] , __lowerCAmelCase : Dict , __lowerCAmelCase : Dict , __lowerCAmelCase : Union[str, Any] ) -> Union[str, Any]:
_A = TFLayoutLMvaModel(config=__lowerCAmelCase )
# text + image
_A = model(__lowerCAmelCase , pixel_values=__lowerCAmelCase , training=__lowerCAmelCase )
_A = model(
__lowerCAmelCase , bbox=__lowerCAmelCase , pixel_values=__lowerCAmelCase , attention_mask=__lowerCAmelCase , token_type_ids=__lowerCAmelCase , training=__lowerCAmelCase , )
_A = model(__lowerCAmelCase , bbox=__lowerCAmelCase , pixel_values=__lowerCAmelCase , training=__lowerCAmelCase )
self.parent.assertEqual(result.last_hidden_state.shape , (self.batch_size, self.seq_length, self.hidden_size) )
# text only
_A = model(__lowerCAmelCase , training=__lowerCAmelCase )
self.parent.assertEqual(
result.last_hidden_state.shape , (self.batch_size, self.text_seq_length, self.hidden_size) )
# image only
_A = model({'''pixel_values''': pixel_values} , training=__lowerCAmelCase )
self.parent.assertEqual(
result.last_hidden_state.shape , (self.batch_size, self.image_seq_length, self.hidden_size) )
def snake_case_ ( self : Tuple , __lowerCAmelCase : List[str] , __lowerCAmelCase : Union[str, Any] , __lowerCAmelCase : Union[str, Any] , __lowerCAmelCase : Any , __lowerCAmelCase : Union[str, Any] , __lowerCAmelCase : Union[str, Any] , __lowerCAmelCase : List[Any] ) -> List[str]:
_A = self.num_labels
_A = TFLayoutLMvaForSequenceClassification(config=__lowerCAmelCase )
_A = model(
__lowerCAmelCase , bbox=__lowerCAmelCase , pixel_values=__lowerCAmelCase , attention_mask=__lowerCAmelCase , token_type_ids=__lowerCAmelCase , labels=__lowerCAmelCase , training=__lowerCAmelCase , )
self.parent.assertEqual(result.logits.shape , (self.batch_size, self.num_labels) )
def snake_case_ ( self : Optional[Any] , __lowerCAmelCase : List[str] , __lowerCAmelCase : Optional[int] , __lowerCAmelCase : List[str] , __lowerCAmelCase : Dict , __lowerCAmelCase : Any , __lowerCAmelCase : Union[str, Any] , __lowerCAmelCase : Dict ) -> List[Any]:
_A = self.num_labels
_A = TFLayoutLMvaForTokenClassification(config=__lowerCAmelCase )
_A = model(
__lowerCAmelCase , bbox=__lowerCAmelCase , pixel_values=__lowerCAmelCase , attention_mask=__lowerCAmelCase , token_type_ids=__lowerCAmelCase , labels=__lowerCAmelCase , training=__lowerCAmelCase , )
self.parent.assertEqual(result.logits.shape , (self.batch_size, self.text_seq_length, self.num_labels) )
def snake_case_ ( self : List[str] , __lowerCAmelCase : List[Any] , __lowerCAmelCase : Any , __lowerCAmelCase : Dict , __lowerCAmelCase : str , __lowerCAmelCase : Tuple , __lowerCAmelCase : Optional[int] , __lowerCAmelCase : Optional[int] ) -> Optional[int]:
_A = 2
_A = TFLayoutLMvaForQuestionAnswering(config=__lowerCAmelCase )
_A = model(
__lowerCAmelCase , bbox=__lowerCAmelCase , pixel_values=__lowerCAmelCase , attention_mask=__lowerCAmelCase , token_type_ids=__lowerCAmelCase , start_positions=__lowerCAmelCase , end_positions=__lowerCAmelCase , training=__lowerCAmelCase , )
self.parent.assertEqual(result.start_logits.shape , (self.batch_size, self.seq_length) )
self.parent.assertEqual(result.end_logits.shape , (self.batch_size, self.seq_length) )
def snake_case_ ( self : Tuple ) -> Union[str, Any]:
_A = self.prepare_config_and_inputs()
((_A) , (_A) , (_A) , (_A) , (_A) , (_A) , (_A) , (_A)) = config_and_inputs
_A = {
'''input_ids''': input_ids,
'''bbox''': bbox,
'''pixel_values''': pixel_values,
'''token_type_ids''': token_type_ids,
'''attention_mask''': input_mask,
}
return config, inputs_dict
@require_tf
class lowerCamelCase__ ( _A , _A , unittest.TestCase):
"""simple docstring"""
a__ : Any = (
(
TFLayoutLMvaModel,
TFLayoutLMvaForQuestionAnswering,
TFLayoutLMvaForSequenceClassification,
TFLayoutLMvaForTokenClassification,
)
if is_tf_available()
else ()
)
a__ : str = (
{"document-question-answering": TFLayoutLMvaForQuestionAnswering, "feature-extraction": TFLayoutLMvaModel}
if is_tf_available()
else {}
)
a__ : Tuple = False
a__ : List[Any] = False
a__ : int = False
def snake_case_ ( self : Dict , __lowerCAmelCase : List[str] , __lowerCAmelCase : Any , __lowerCAmelCase : Optional[Any] , __lowerCAmelCase : List[Any] , __lowerCAmelCase : Any ) -> Union[str, Any]:
return True
def snake_case_ ( self : Optional[Any] , __lowerCAmelCase : Tuple , __lowerCAmelCase : Optional[Any] , __lowerCAmelCase : List[Any]=False ) -> dict:
_A = copy.deepcopy(__lowerCAmelCase )
if model_class in get_values(__lowerCAmelCase ):
_A = {
k: tf.tile(tf.expand_dims(__lowerCAmelCase , 1 ) , (1, self.model_tester.num_choices) + (1,) * (v.ndim - 1) )
if isinstance(__lowerCAmelCase , tf.Tensor ) and v.ndim > 0
else v
for k, v in inputs_dict.items()
}
if return_labels:
if model_class in get_values(__lowerCAmelCase ):
_A = tf.ones(self.model_tester.batch_size , dtype=tf.intaa )
elif model_class in get_values(__lowerCAmelCase ):
_A = tf.zeros(self.model_tester.batch_size , dtype=tf.intaa )
_A = tf.zeros(self.model_tester.batch_size , dtype=tf.intaa )
elif model_class in get_values(__lowerCAmelCase ):
_A = tf.zeros(self.model_tester.batch_size , dtype=tf.intaa )
elif model_class in get_values(__lowerCAmelCase ):
_A = tf.zeros(
(self.model_tester.batch_size, self.model_tester.text_seq_length) , dtype=tf.intaa )
return inputs_dict
def snake_case_ ( self : Any ) -> List[str]:
_A = TFLayoutLMvaModelTester(self )
_A = ConfigTester(self , config_class=__lowerCAmelCase , hidden_size=37 )
def snake_case_ ( self : Dict ) -> Dict:
self.config_tester.run_common_tests()
def snake_case_ ( self : str ) -> int:
_A , _A = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
_A = model_class(__lowerCAmelCase )
if getattr(__lowerCAmelCase , '''hf_compute_loss''' , __lowerCAmelCase ):
# The number of elements in the loss should be the same as the number of elements in the label
_A = self._prepare_for_class(inputs_dict.copy() , __lowerCAmelCase , return_labels=__lowerCAmelCase )
_A = prepared_for_class[
sorted(prepared_for_class.keys() - inputs_dict.keys() , reverse=__lowerCAmelCase )[0]
]
_A = added_label.shape.as_list()[:1]
# Test that model correctly compute the loss with kwargs
_A = self._prepare_for_class(inputs_dict.copy() , __lowerCAmelCase , return_labels=__lowerCAmelCase )
_A = prepared_for_class.pop('''input_ids''' )
_A = model(__lowerCAmelCase , **__lowerCAmelCase )[0]
self.assertTrue(loss.shape.as_list() == expected_loss_size or loss.shape.as_list() == [1] )
# Test that model correctly compute the loss when we mask some positions
_A = self._prepare_for_class(inputs_dict.copy() , __lowerCAmelCase , return_labels=__lowerCAmelCase )
_A = prepared_for_class.pop('''input_ids''' )
if "labels" in prepared_for_class:
_A = prepared_for_class['''labels'''].numpy()
if len(labels.shape ) > 1 and labels.shape[1] != 1:
_A = -1_00
_A = tf.convert_to_tensor(__lowerCAmelCase )
_A = model(__lowerCAmelCase , **__lowerCAmelCase )[0]
self.assertTrue(loss.shape.as_list() == expected_loss_size or loss.shape.as_list() == [1] )
self.assertTrue(not np.any(np.isnan(loss.numpy() ) ) )
# Test that model correctly compute the loss with a dict
_A = self._prepare_for_class(inputs_dict.copy() , __lowerCAmelCase , return_labels=__lowerCAmelCase )
_A = model(__lowerCAmelCase )[0]
self.assertTrue(loss.shape.as_list() == expected_loss_size or loss.shape.as_list() == [1] )
# Test that model correctly compute the loss with a tuple
_A = self._prepare_for_class(inputs_dict.copy() , __lowerCAmelCase , return_labels=__lowerCAmelCase )
# Get keys that were added with the _prepare_for_class function
_A = prepared_for_class.keys() - inputs_dict.keys()
_A = inspect.signature(model.call ).parameters
_A = list(signature.keys() )
# Create a dictionary holding the location of the tensors in the tuple
_A = {0: '''input_ids'''}
for label_key in label_keys:
_A = signature_names.index(__lowerCAmelCase )
_A = label_key
_A = sorted(tuple_index_mapping.items() )
# Initialize a list with their default values, update the values and convert to a tuple
_A = []
for name in signature_names:
if name != "kwargs":
list_input.append(signature[name].default )
for index, value in sorted_tuple_index_mapping:
_A = prepared_for_class[value]
_A = tuple(__lowerCAmelCase )
# Send to model
_A = model(tuple_input[:-1] )[0]
self.assertTrue(loss.shape.as_list() == expected_loss_size or loss.shape.as_list() == [1] )
def snake_case_ ( self : Tuple ) -> Optional[int]:
(
(
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) ,
) = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(__lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase )
def snake_case_ ( self : Dict ) -> int:
(
(
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) ,
) = self.model_tester.prepare_config_and_inputs()
for type in ["absolute", "relative_key", "relative_key_query"]:
_A = type
self.model_tester.create_and_check_model(__lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase )
def snake_case_ ( self : List[Any] ) -> Dict:
(
(
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) ,
) = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_sequence_classification(
__lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase )
def snake_case_ ( self : Dict ) -> List[str]:
(
(
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) ,
) = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_token_classification(
__lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase )
def snake_case_ ( self : List[Any] ) -> Optional[Any]:
(
(
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) , (
_A
) ,
) = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_question_answering(
__lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase )
@slow
def snake_case_ ( self : Dict ) -> Union[str, Any]:
for model_name in TF_LAYOUTLMV3_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
_A = TFLayoutLMvaModel.from_pretrained(__lowerCAmelCase )
self.assertIsNotNone(__lowerCAmelCase )
def SCREAMING_SNAKE_CASE_ ( ) -> Optional[Any]:
_A = Image.open('''./tests/fixtures/tests_samples/COCO/000000039769.png''' )
return image
@require_tf
class lowerCamelCase__ ( unittest.TestCase):
"""simple docstring"""
@cached_property
def snake_case_ ( self : List[Any] ) -> Union[str, Any]:
return LayoutLMvaImageProcessor(apply_ocr=__lowerCAmelCase ) if is_vision_available() else None
@slow
def snake_case_ ( self : Union[str, Any] ) -> Any:
_A = TFLayoutLMvaModel.from_pretrained('''microsoft/layoutlmv3-base''' )
_A = self.default_image_processor
_A = prepare_img()
_A = image_processor(images=__lowerCAmelCase , return_tensors='''tf''' ).pixel_values
_A = tf.constant([[1, 2]] )
_A = tf.expand_dims(tf.constant([[1, 2, 3, 4], [5, 6, 7, 8]] ) , axis=0 )
# forward pass
_A = model(input_ids=__lowerCAmelCase , bbox=__lowerCAmelCase , pixel_values=__lowerCAmelCase , training=__lowerCAmelCase )
# verify the logits
_A = (1, 1_99, 7_68)
self.assertEqual(outputs.last_hidden_state.shape , __lowerCAmelCase )
_A = tf.constant(
[[-0.0529, 0.3618, 0.1632], [-0.1587, -0.1667, -0.0400], [-0.1557, -0.1671, -0.0505]] )
self.assertTrue(np.allclose(outputs.last_hidden_state[0, :3, :3] , __lowerCAmelCase , atol=1E-4 ) )
| 2 |
import torch
from diffusers import DDPMParallelScheduler
from .test_schedulers import SchedulerCommonTest
class _a ( snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Optional[Any] = (DDPMParallelScheduler,)
def __A ( self : List[Any] , **UpperCAmelCase : Optional[int] ):
A_ = {
"num_train_timesteps": 1000,
"beta_start": 0.0_001,
"beta_end": 0.02,
"beta_schedule": "linear",
"variance_type": "fixed_small",
"clip_sample": True,
}
config.update(**UpperCAmelCase )
return config
def __A ( self : Optional[Any] ):
for timesteps in [1, 5, 100, 1000]:
self.check_over_configs(num_train_timesteps=UpperCAmelCase )
def __A ( self : Dict ):
for beta_start, beta_end in zip([0.0_001, 0.001, 0.01, 0.1] , [0.002, 0.02, 0.2, 2] ):
self.check_over_configs(beta_start=UpperCAmelCase , beta_end=UpperCAmelCase )
def __A ( self : int ):
for schedule in ["linear", "squaredcos_cap_v2"]:
self.check_over_configs(beta_schedule=UpperCAmelCase )
def __A ( self : Tuple ):
for variance in ["fixed_small", "fixed_large", "other"]:
self.check_over_configs(variance_type=UpperCAmelCase )
def __A ( self : int ):
for clip_sample in [True, False]:
self.check_over_configs(clip_sample=UpperCAmelCase )
def __A ( self : Union[str, Any] ):
self.check_over_configs(thresholding=UpperCAmelCase )
for threshold in [0.5, 1.0, 2.0]:
for prediction_type in ["epsilon", "sample", "v_prediction"]:
self.check_over_configs(
thresholding=UpperCAmelCase , prediction_type=UpperCAmelCase , sample_max_value=UpperCAmelCase , )
def __A ( self : Optional[int] ):
for prediction_type in ["epsilon", "sample", "v_prediction"]:
self.check_over_configs(prediction_type=UpperCAmelCase )
def __A ( self : Tuple ):
for t in [0, 500, 999]:
self.check_over_forward(time_step=UpperCAmelCase )
def __A ( self : Tuple ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
assert torch.sum(torch.abs(scheduler._get_variance(0 ) - 0.0 ) ) < 1E-5
assert torch.sum(torch.abs(scheduler._get_variance(487 ) - 0.00_979 ) ) < 1E-5
assert torch.sum(torch.abs(scheduler._get_variance(999 ) - 0.02 ) ) < 1E-5
def __A ( self : List[Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = len(UpperCAmelCase )
A_ = self.dummy_model()
A_ = self.dummy_sample_deter
A_ = self.dummy_sample_deter + 0.1
A_ = self.dummy_sample_deter - 0.1
A_ = samplea.shape[0]
A_ = torch.stack([samplea, samplea, samplea] , dim=0 )
A_ = torch.arange(UpperCAmelCase )[0:3, None].repeat(1 , UpperCAmelCase )
A_ = model(samples.flatten(0 , 1 ) , timesteps.flatten(0 , 1 ) )
A_ = scheduler.batch_step_no_noise(UpperCAmelCase , timesteps.flatten(0 , 1 ) , samples.flatten(0 , 1 ) )
A_ = torch.sum(torch.abs(UpperCAmelCase ) )
A_ = torch.mean(torch.abs(UpperCAmelCase ) )
assert abs(result_sum.item() - 1_153.1_833 ) < 1E-2
assert abs(result_mean.item() - 0.5_005 ) < 1E-3
def __A ( self : Tuple ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = len(UpperCAmelCase )
A_ = self.dummy_model()
A_ = self.dummy_sample_deter
A_ = torch.manual_seed(0 )
for t in reversed(range(UpperCAmelCase ) ):
# 1. predict noise residual
A_ = model(UpperCAmelCase , UpperCAmelCase )
# 2. predict previous mean of sample x_t-1
A_ = scheduler.step(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase , generator=UpperCAmelCase ).prev_sample
A_ = pred_prev_sample
A_ = torch.sum(torch.abs(UpperCAmelCase ) )
A_ = torch.mean(torch.abs(UpperCAmelCase ) )
assert abs(result_sum.item() - 258.9_606 ) < 1E-2
assert abs(result_mean.item() - 0.3_372 ) < 1E-3
def __A ( self : Tuple ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config(prediction_type="v_prediction" )
A_ = scheduler_class(**UpperCAmelCase )
A_ = len(UpperCAmelCase )
A_ = self.dummy_model()
A_ = self.dummy_sample_deter
A_ = torch.manual_seed(0 )
for t in reversed(range(UpperCAmelCase ) ):
# 1. predict noise residual
A_ = model(UpperCAmelCase , UpperCAmelCase )
# 2. predict previous mean of sample x_t-1
A_ = scheduler.step(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase , generator=UpperCAmelCase ).prev_sample
A_ = pred_prev_sample
A_ = torch.sum(torch.abs(UpperCAmelCase ) )
A_ = torch.mean(torch.abs(UpperCAmelCase ) )
assert abs(result_sum.item() - 202.0_296 ) < 1E-2
assert abs(result_mean.item() - 0.2_631 ) < 1E-3
def __A ( self : Union[str, Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = [100, 87, 50, 1, 0]
scheduler.set_timesteps(timesteps=UpperCAmelCase )
A_ = scheduler.timesteps
for i, timestep in enumerate(UpperCAmelCase ):
if i == len(UpperCAmelCase ) - 1:
A_ = -1
else:
A_ = timesteps[i + 1]
A_ = scheduler.previous_timestep(UpperCAmelCase )
A_ = prev_t.item()
self.assertEqual(UpperCAmelCase , UpperCAmelCase )
def __A ( self : List[Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = [100, 87, 50, 51, 0]
with self.assertRaises(UpperCAmelCase , msg="`custom_timesteps` must be in descending order." ):
scheduler.set_timesteps(timesteps=UpperCAmelCase )
def __A ( self : List[Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = [100, 87, 50, 1, 0]
A_ = len(UpperCAmelCase )
with self.assertRaises(UpperCAmelCase , msg="Can only pass one of `num_inference_steps` or `custom_timesteps`." ):
scheduler.set_timesteps(num_inference_steps=UpperCAmelCase , timesteps=UpperCAmelCase )
def __A ( self : Optional[Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = [scheduler.config.num_train_timesteps]
with self.assertRaises(
UpperCAmelCase , msg="`timesteps` must start before `self.config.train_timesteps`: {scheduler.config.num_train_timesteps}}" , ):
scheduler.set_timesteps(timesteps=UpperCAmelCase ) | 86 | 0 |
'''simple docstring'''
from typing import TYPE_CHECKING
from ...utils import (
OptionalDependencyNotAvailable,
_LazyModule,
is_flax_available,
is_sentencepiece_available,
is_tf_available,
is_tokenizers_available,
is_torch_available,
)
if is_sentencepiece_available():
from ..ta.tokenization_ta import TaTokenizer
else:
from ...utils.dummy_sentencepiece_objects import TaTokenizer
lowerCAmelCase : Tuple = TaTokenizer
if is_tokenizers_available():
from ..ta.tokenization_ta_fast import TaTokenizerFast
else:
from ...utils.dummy_tokenizers_objects import TaTokenizerFast
lowerCAmelCase : Optional[int] = TaTokenizerFast
lowerCAmelCase : Any = {'configuration_mt5': ['MT5Config', 'MT5OnnxConfig']}
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
lowerCAmelCase : Optional[int] = [
'MT5EncoderModel',
'MT5ForConditionalGeneration',
'MT5ForQuestionAnswering',
'MT5Model',
'MT5PreTrainedModel',
'MT5Stack',
]
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
lowerCAmelCase : Dict = ['TFMT5EncoderModel', 'TFMT5ForConditionalGeneration', 'TFMT5Model']
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
lowerCAmelCase : Optional[Any] = ['FlaxMT5EncoderModel', 'FlaxMT5ForConditionalGeneration', 'FlaxMT5Model']
if TYPE_CHECKING:
from .configuration_mta import MTaConfig, MTaOnnxConfig
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_mta import (
MTaEncoderModel,
MTaForConditionalGeneration,
MTaForQuestionAnswering,
MTaModel,
MTaPreTrainedModel,
MTaStack,
)
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tf_mta import TFMTaEncoderModel, TFMTaForConditionalGeneration, TFMTaModel
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_flax_mta import FlaxMTaEncoderModel, FlaxMTaForConditionalGeneration, FlaxMTaModel
else:
import sys
lowerCAmelCase : Tuple = _LazyModule(
__name__,
globals()['__file__'],
_import_structure,
extra_objects={'MT5Tokenizer': MTaTokenizer, 'MT5TokenizerFast': MTaTokenizerFast},
module_spec=__spec__,
)
| 3 |
import argparse
import json
import os
import torch
from transformers import LukeConfig, LukeModel, LukeTokenizer, RobertaTokenizer
from transformers.tokenization_utils_base import AddedToken
@torch.no_grad()
def __snake_case ( __UpperCamelCase : Tuple ,__UpperCamelCase : Dict ,__UpperCamelCase : Union[str, Any] ,__UpperCamelCase : List[Any] ,__UpperCamelCase : List[Any] ):
"""simple docstring"""
with open(__UpperCamelCase ) as metadata_file:
A_ = json.load(__UpperCamelCase )
A_ = LukeConfig(use_entity_aware_attention=__UpperCamelCase ,**metadata["model_config"] )
# Load in the weights from the checkpoint_path
A_ = torch.load(__UpperCamelCase ,map_location="cpu" )
# Load the entity vocab file
A_ = load_entity_vocab(__UpperCamelCase )
A_ = RobertaTokenizer.from_pretrained(metadata["model_config"]["bert_model_name"] )
# Add special tokens to the token vocabulary for downstream tasks
A_ = AddedToken("<ent>" ,lstrip=__UpperCamelCase ,rstrip=__UpperCamelCase )
A_ = AddedToken("<ent2>" ,lstrip=__UpperCamelCase ,rstrip=__UpperCamelCase )
tokenizer.add_special_tokens({"additional_special_tokens": [entity_token_a, entity_token_a]} )
config.vocab_size += 2
print(f'''Saving tokenizer to {pytorch_dump_folder_path}''' )
tokenizer.save_pretrained(__UpperCamelCase )
with open(os.path.join(__UpperCamelCase ,LukeTokenizer.vocab_files_names["entity_vocab_file"] ) ,"w" ) as f:
json.dump(__UpperCamelCase ,__UpperCamelCase )
A_ = LukeTokenizer.from_pretrained(__UpperCamelCase )
# Initialize the embeddings of the special tokens
A_ = state_dict["embeddings.word_embeddings.weight"]
A_ = word_emb[tokenizer.convert_tokens_to_ids(["@"] )[0]].unsqueeze(0 )
A_ = word_emb[tokenizer.convert_tokens_to_ids(["#"] )[0]].unsqueeze(0 )
A_ = torch.cat([word_emb, ent_emb, enta_emb] )
# Initialize the query layers of the entity-aware self-attention mechanism
for layer_index in range(config.num_hidden_layers ):
for matrix_name in ["query.weight", "query.bias"]:
A_ = f'''encoder.layer.{layer_index}.attention.self.'''
A_ = state_dict[prefix + matrix_name]
A_ = state_dict[prefix + matrix_name]
A_ = state_dict[prefix + matrix_name]
# Initialize the embedding of the [MASK2] entity using that of the [MASK] entity for downstream tasks
A_ = state_dict["entity_embeddings.entity_embeddings.weight"]
A_ = entity_emb[entity_vocab["[MASK]"]]
A_ = LukeModel(config=__UpperCamelCase ).eval()
A_ , A_ = model.load_state_dict(__UpperCamelCase ,strict=__UpperCamelCase )
if not (len(__UpperCamelCase ) == 1 and missing_keys[0] == "embeddings.position_ids"):
raise ValueError(f'''Missing keys {", ".join(__UpperCamelCase )}. Expected only missing embeddings.position_ids''' )
if not (all(key.startswith("entity_predictions" ) or key.startswith("lm_head" ) for key in unexpected_keys )):
raise ValueError(
"Unexpected keys"
f''' {", ".join([key for key in unexpected_keys if not (key.startswith("entity_predictions" ) or key.startswith("lm_head" ))] )}''' )
# Check outputs
A_ = LukeTokenizer.from_pretrained(__UpperCamelCase ,task="entity_classification" )
A_ = (
"Top seed Ana Ivanovic said on Thursday she could hardly believe her luck as a fortuitous netcord helped the"
" new world number one avoid a humiliating second- round exit at Wimbledon ."
)
A_ = (39, 42)
A_ = tokenizer(__UpperCamelCase ,entity_spans=[span] ,add_prefix_space=__UpperCamelCase ,return_tensors="pt" )
A_ = model(**__UpperCamelCase )
# Verify word hidden states
if model_size == "large":
A_ = torch.Size((1, 42, 1024) )
A_ = torch.tensor(
[[0.0133, 0.0865, 0.0095], [0.3093, -0.2576, -0.7418], [-0.1720, -0.2117, -0.2869]] )
else: # base
A_ = torch.Size((1, 42, 768) )
A_ = torch.tensor([[0.0037, 0.1368, -0.0091], [0.1099, 0.3329, -0.1095], [0.0765, 0.5335, 0.1179]] )
if not (outputs.last_hidden_state.shape == expected_shape):
raise ValueError(
f'''Outputs.last_hidden_state.shape is {outputs.last_hidden_state.shape}, Expected shape is {expected_shape}''' )
if not torch.allclose(outputs.last_hidden_state[0, :3, :3] ,__UpperCamelCase ,atol=1E-4 ):
raise ValueError
# Verify entity hidden states
if model_size == "large":
A_ = torch.Size((1, 1, 1024) )
A_ = torch.tensor([[0.0466, -0.0106, -0.0179]] )
else: # base
A_ = torch.Size((1, 1, 768) )
A_ = torch.tensor([[0.1457, 0.1044, 0.0174]] )
if not (outputs.entity_last_hidden_state.shape != expected_shape):
raise ValueError(
f'''Outputs.entity_last_hidden_state.shape is {outputs.entity_last_hidden_state.shape}, Expected shape is'''
f''' {expected_shape}''' )
if not torch.allclose(outputs.entity_last_hidden_state[0, :3, :3] ,__UpperCamelCase ,atol=1E-4 ):
raise ValueError
# Finally, save our PyTorch model and tokenizer
print("Saving PyTorch model to {}".format(__UpperCamelCase ) )
model.save_pretrained(__UpperCamelCase )
def __snake_case ( __UpperCamelCase : str ):
"""simple docstring"""
A_ = {}
with open(__UpperCamelCase ,"r" ,encoding="utf-8" ) as f:
for index, line in enumerate(__UpperCamelCase ):
A_ , A_ = line.rstrip().split("\t" )
A_ = index
return entity_vocab
if __name__ == "__main__":
__a :Dict = argparse.ArgumentParser()
# Required parameters
parser.add_argument('--checkpoint_path', type=str, help='Path to a pytorch_model.bin file.')
parser.add_argument(
'--metadata_path', default=None, type=str, help='Path to a metadata.json file, defining the configuration.'
)
parser.add_argument(
'--entity_vocab_path',
default=None,
type=str,
help='Path to an entity_vocab.tsv file, containing the entity vocabulary.',
)
parser.add_argument(
'--pytorch_dump_folder_path', default=None, type=str, help='Path to where to dump the output PyTorch model.'
)
parser.add_argument(
'--model_size', default='base', type=str, choices=['base', 'large'], help='Size of the model to be converted.'
)
__a :Tuple = parser.parse_args()
convert_luke_checkpoint(
args.checkpoint_path,
args.metadata_path,
args.entity_vocab_path,
args.pytorch_dump_folder_path,
args.model_size,
) | 86 | 0 |
"""simple docstring"""
from typing import List, Union
from ..utils import (
add_end_docstrings,
is_tf_available,
is_torch_available,
is_vision_available,
logging,
requires_backends,
)
from .base import PIPELINE_INIT_ARGS, Pipeline
if is_vision_available():
from PIL import Image
from ..image_utils import load_image
if is_tf_available():
from ..models.auto.modeling_tf_auto import TF_MODEL_FOR_VISION_2_SEQ_MAPPING
if is_torch_available():
import torch
from ..models.auto.modeling_auto import MODEL_FOR_VISION_2_SEQ_MAPPING
__UpperCamelCase : Union[str, Any] = logging.get_logger(__name__)
@add_end_docstrings(a__ )
class a ( a__ ):
def __init__( self , *_snake_case , **_snake_case ):
"""simple docstring"""
super().__init__(*_snake_case , **_snake_case )
requires_backends(self , 'vision' )
self.check_model_type(
TF_MODEL_FOR_VISION_2_SEQ_MAPPING if self.framework == 'tf' else MODEL_FOR_VISION_2_SEQ_MAPPING )
def UpperCamelCase__ ( self , _snake_case=None , _snake_case=None , _snake_case=None ):
"""simple docstring"""
lowerCAmelCase = {}
lowerCAmelCase = {}
if prompt is not None:
lowerCAmelCase = prompt
if generate_kwargs is not None:
lowerCAmelCase = generate_kwargs
if max_new_tokens is not None:
if "generate_kwargs" not in forward_kwargs:
lowerCAmelCase = {}
if "max_new_tokens" in forward_kwargs["generate_kwargs"]:
raise ValueError(
'\'max_new_tokens\' is defined twice, once in \'generate_kwargs\' and once as a direct parameter,'
' please use only one' )
lowerCAmelCase = max_new_tokens
return preprocess_params, forward_kwargs, {}
def __call__( self , _snake_case , **_snake_case ):
"""simple docstring"""
return super().__call__(_snake_case , **_snake_case )
def UpperCamelCase__ ( self , _snake_case , _snake_case=None ):
"""simple docstring"""
lowerCAmelCase = load_image(_snake_case )
if prompt is not None:
if not isinstance(_snake_case , _snake_case ):
raise ValueError(
F'Received an invalid text input, got - {type(_snake_case )} - but expected a single string. '
'Note also that one single text can be provided for conditional image to text generation.' )
lowerCAmelCase = self.model.config.model_type
if model_type == "git":
lowerCAmelCase = self.image_processor(images=_snake_case , return_tensors=self.framework )
lowerCAmelCase = self.tokenizer(text=_snake_case , add_special_tokens=_snake_case ).input_ids
lowerCAmelCase = [self.tokenizer.cls_token_id] + input_ids
lowerCAmelCase = torch.tensor(_snake_case ).unsqueeze(0 )
model_inputs.update({'input_ids': input_ids} )
elif model_type == "pix2struct":
lowerCAmelCase = self.image_processor(images=_snake_case , header_text=_snake_case , return_tensors=self.framework )
elif model_type != "vision-encoder-decoder":
# vision-encoder-decoder does not support conditional generation
lowerCAmelCase = self.image_processor(images=_snake_case , return_tensors=self.framework )
lowerCAmelCase = self.tokenizer(_snake_case , return_tensors=self.framework )
model_inputs.update(_snake_case )
else:
raise ValueError(F'Model type {model_type} does not support conditional text generation' )
else:
lowerCAmelCase = self.image_processor(images=_snake_case , return_tensors=self.framework )
if self.model.config.model_type == "git" and prompt is None:
lowerCAmelCase = None
return model_inputs
def UpperCamelCase__ ( self , _snake_case , _snake_case=None ):
"""simple docstring"""
if (
"input_ids" in model_inputs
and isinstance(model_inputs['input_ids'] , _snake_case )
and all(x is None for x in model_inputs['input_ids'] )
):
lowerCAmelCase = None
if generate_kwargs is None:
lowerCAmelCase = {}
# FIXME: We need to pop here due to a difference in how `generation.py` and `generation.tf_utils.py`
# parse inputs. In the Tensorflow version, `generate` raises an error if we don't use `input_ids` whereas
# the PyTorch version matches it with `self.model.main_input_name` or `self.model.encoder.main_input_name`
# in the `_prepare_model_inputs` method.
lowerCAmelCase = model_inputs.pop(self.model.main_input_name )
lowerCAmelCase = self.model.generate(_snake_case , **_snake_case , **_snake_case )
return model_outputs
def UpperCamelCase__ ( self , _snake_case ):
"""simple docstring"""
lowerCAmelCase = []
for output_ids in model_outputs:
lowerCAmelCase = {
'generated_text': self.tokenizer.decode(
_snake_case , skip_special_tokens=_snake_case , )
}
records.append(_snake_case )
return records
| 4 |
import math
import os
from copy import deepcopy
import datasets
import evaluate
import torch
import transformers
from datasets import load_dataset
from torch.utils.data import DataLoader
from transformers import AutoModelForSequenceClassification, AutoTokenizer
from accelerate import Accelerator
from accelerate.test_utils import RegressionDataset, RegressionModel
from accelerate.utils import is_tpu_available, set_seed
__a :Optional[Any] = 'true'
def __snake_case ( __UpperCamelCase : Any ,__UpperCamelCase : List[Any]=82 ,__UpperCamelCase : Dict=16 ):
"""simple docstring"""
set_seed(42 )
A_ = RegressionModel()
A_ = deepcopy(__UpperCamelCase )
A_ = RegressionDataset(length=__UpperCamelCase )
A_ = DataLoader(__UpperCamelCase ,batch_size=__UpperCamelCase )
model.to(accelerator.device )
A_ , A_ = accelerator.prepare(__UpperCamelCase ,__UpperCamelCase )
return model, ddp_model, dataloader
def __snake_case ( __UpperCamelCase : Accelerator ,__UpperCamelCase : Dict=False ):
"""simple docstring"""
A_ = AutoTokenizer.from_pretrained("hf-internal-testing/mrpc-bert-base-cased" )
A_ = load_dataset("glue" ,"mrpc" ,split="validation" )
def tokenize_function(__UpperCamelCase : Optional[Any] ):
A_ = tokenizer(examples["sentence1"] ,examples["sentence2"] ,truncation=__UpperCamelCase ,max_length=__UpperCamelCase )
return outputs
with accelerator.main_process_first():
A_ = dataset.map(
__UpperCamelCase ,batched=__UpperCamelCase ,remove_columns=["idx", "sentence1", "sentence2"] ,)
A_ = tokenized_datasets.rename_column("label" ,"labels" )
def collate_fn(__UpperCamelCase : Union[str, Any] ):
if use_longest:
return tokenizer.pad(__UpperCamelCase ,padding="longest" ,return_tensors="pt" )
return tokenizer.pad(__UpperCamelCase ,padding="max_length" ,max_length=128 ,return_tensors="pt" )
return DataLoader(__UpperCamelCase ,shuffle=__UpperCamelCase ,collate_fn=__UpperCamelCase ,batch_size=16 )
def __snake_case ( __UpperCamelCase : Optional[Any] ,__UpperCamelCase : str ):
"""simple docstring"""
A_ = Accelerator(dispatch_batches=__UpperCamelCase ,split_batches=__UpperCamelCase )
A_ = get_dataloader(__UpperCamelCase ,not dispatch_batches )
A_ = AutoModelForSequenceClassification.from_pretrained(
"hf-internal-testing/mrpc-bert-base-cased" ,return_dict=__UpperCamelCase )
A_ , A_ = accelerator.prepare(__UpperCamelCase ,__UpperCamelCase )
return {"ddp": [ddp_model, ddp_dataloader, "cuda:0"], "no": [model, dataloader, accelerator.device]}, accelerator
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : int ,__UpperCamelCase : Optional[Any] ):
"""simple docstring"""
A_ = []
for batch in dataloader:
A_ , A_ = batch.values()
with torch.no_grad():
A_ = model(__UpperCamelCase )
A_ , A_ = accelerator.gather_for_metrics((logit, target) )
logits_and_targets.append((logit, target) )
A_ , A_ = [], []
for logit, targ in logits_and_targets:
logits.append(__UpperCamelCase )
targs.append(__UpperCamelCase )
A_ , A_ = torch.cat(__UpperCamelCase ), torch.cat(__UpperCamelCase )
return logits, targs
def __snake_case ( __UpperCamelCase : Accelerator ,__UpperCamelCase : Dict=82 ,__UpperCamelCase : List[Any]=False ,__UpperCamelCase : Dict=False ,__UpperCamelCase : Optional[int]=16 ):
"""simple docstring"""
A_ , A_ , A_ = get_basic_setup(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
A_ , A_ = generate_predictions(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
assert (
len(__UpperCamelCase ) == num_samples
), f'''Unexpected number of inputs:\n Expected: {num_samples}\n Actual: {len(__UpperCamelCase )}'''
def __snake_case ( __UpperCamelCase : bool = False ,__UpperCamelCase : bool = False ):
"""simple docstring"""
A_ = evaluate.load("glue" ,"mrpc" )
A_ , A_ = get_mrpc_setup(__UpperCamelCase ,__UpperCamelCase )
# First do baseline
A_ , A_ , A_ = setup["no"]
model.to(__UpperCamelCase )
model.eval()
for batch in dataloader:
batch.to(__UpperCamelCase )
with torch.inference_mode():
A_ = model(**__UpperCamelCase )
A_ = outputs.logits.argmax(dim=-1 )
metric.add_batch(predictions=__UpperCamelCase ,references=batch["labels"] )
A_ = metric.compute()
# Then do distributed
A_ , A_ , A_ = setup["ddp"]
model.eval()
for batch in dataloader:
with torch.inference_mode():
A_ = model(**__UpperCamelCase )
A_ = outputs.logits.argmax(dim=-1 )
A_ = batch["labels"]
A_ , A_ = accelerator.gather_for_metrics((preds, references) )
metric.add_batch(predictions=__UpperCamelCase ,references=__UpperCamelCase )
A_ = metric.compute()
for key in "accuracy f1".split():
assert math.isclose(
baseline[key] ,distributed[key] ), f'''Baseline and Distributed are not the same for key {key}:\n\tBaseline: {baseline[key]}\n\tDistributed: {distributed[key]}\n'''
def __snake_case ( ):
"""simple docstring"""
A_ = Accelerator(split_batches=__UpperCamelCase ,dispatch_batches=__UpperCamelCase )
if accelerator.is_local_main_process:
datasets.utils.logging.set_verbosity_warning()
transformers.utils.logging.set_verbosity_warning()
else:
datasets.utils.logging.set_verbosity_error()
transformers.utils.logging.set_verbosity_error()
# These are a bit slower so they should only be ran on the GPU or TPU
if torch.cuda.is_available() or is_tpu_available():
if accelerator.is_local_main_process:
print("**Testing gather_for_metrics**" )
for split_batches in [True, False]:
for dispatch_batches in [True, False]:
if accelerator.is_local_main_process:
print(f'''With: `split_batches={split_batches}`, `dispatch_batches={dispatch_batches}`''' )
test_mrpc(__UpperCamelCase ,__UpperCamelCase )
accelerator.state._reset_state()
if accelerator.is_local_main_process:
print("**Test torch metrics**" )
for split_batches in [True, False]:
for dispatch_batches in [True, False]:
A_ = Accelerator(split_batches=__UpperCamelCase ,dispatch_batches=__UpperCamelCase )
if accelerator.is_local_main_process:
print(f'''With: `split_batches={split_batches}`, `dispatch_batches={dispatch_batches}`, length=99''' )
test_torch_metrics(__UpperCamelCase ,99 )
accelerator.state._reset_state()
if accelerator.is_local_main_process:
print("**Test last batch is not dropped when perfectly divisible**" )
A_ = Accelerator()
test_torch_metrics(__UpperCamelCase ,512 )
accelerator.state._reset_state()
def __snake_case ( __UpperCamelCase : List[Any] ):
"""simple docstring"""
main()
if __name__ == "__main__":
main() | 86 | 0 |
'''simple docstring'''
from __future__ import annotations
def A (__lowerCamelCase :list[int | float] , __lowerCamelCase :int , __lowerCamelCase :int ):
if len(__lowerCamelCase ) == 0:
raise ValueError("""find_max() arg is an empty sequence""" )
if (
left >= len(__lowerCamelCase )
or left < -len(__lowerCamelCase )
or right >= len(__lowerCamelCase )
or right < -len(__lowerCamelCase )
):
raise IndexError("""list index out of range""" )
if left == right:
return nums[left]
_lowerCAmelCase = (left + right) >> 1 # the middle
_lowerCAmelCase = find_max(__lowerCamelCase , __lowerCamelCase , __lowerCamelCase ) # find max in range[left, mid]
_lowerCAmelCase = find_max(__lowerCamelCase , mid + 1 , __lowerCamelCase ) # find max in range[mid + 1, right]
return left_max if left_max >= right_max else right_max
if __name__ == "__main__":
import doctest
doctest.testmod(verbose=True)
| 5 |
import argparse
import os
from transformers.utils import direct_transformers_import
# All paths are set with the intent you should run this script from the root of the repo with the command
# python utils/check_task_guides.py
__a :Optional[Any] = 'src/transformers'
__a :Tuple = 'docs/source/en/tasks'
def __snake_case ( __UpperCamelCase : List[Any] ,__UpperCamelCase : Dict ,__UpperCamelCase : int ):
"""simple docstring"""
with open(__UpperCamelCase ,"r" ,encoding="utf-8" ,newline="\n" ) as f:
A_ = f.readlines()
# Find the start prompt.
A_ = 0
while not lines[start_index].startswith(__UpperCamelCase ):
start_index += 1
start_index += 1
A_ = start_index
while not lines[end_index].startswith(__UpperCamelCase ):
end_index += 1
end_index -= 1
while len(lines[start_index] ) <= 1:
start_index += 1
while len(lines[end_index] ) <= 1:
end_index -= 1
end_index += 1
return "".join(lines[start_index:end_index] ), start_index, end_index, lines
# This is to make sure the transformers module imported is the one in the repo.
__a :List[str] = direct_transformers_import(TRANSFORMERS_PATH)
__a :Optional[Any] = {
'asr.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_CTC_MAPPING_NAMES,
'audio_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_AUDIO_CLASSIFICATION_MAPPING_NAMES,
'language_modeling.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_CAUSAL_LM_MAPPING_NAMES,
'image_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_IMAGE_CLASSIFICATION_MAPPING_NAMES,
'masked_language_modeling.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_MASKED_LM_MAPPING_NAMES,
'multiple_choice.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_MULTIPLE_CHOICE_MAPPING_NAMES,
'object_detection.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_OBJECT_DETECTION_MAPPING_NAMES,
'question_answering.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_QUESTION_ANSWERING_MAPPING_NAMES,
'semantic_segmentation.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEMANTIC_SEGMENTATION_MAPPING_NAMES,
'sequence_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING_NAMES,
'summarization.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES,
'token_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING_NAMES,
'translation.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES,
'video_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_VIDEO_CLASSIFICATION_MAPPING_NAMES,
'document_question_answering.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_DOCUMENT_QUESTION_ANSWERING_MAPPING_NAMES,
'monocular_depth_estimation.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_DEPTH_ESTIMATION_MAPPING_NAMES,
}
# This list contains model types used in some task guides that are not in `CONFIG_MAPPING_NAMES` (therefore not in any
# `MODEL_MAPPING_NAMES` or any `MODEL_FOR_XXX_MAPPING_NAMES`).
__a :Optional[Any] = {
'summarization.md': ('nllb',),
'translation.md': ('nllb',),
}
def __snake_case ( __UpperCamelCase : Tuple ):
"""simple docstring"""
A_ = TASK_GUIDE_TO_MODELS[task_guide]
A_ = SPECIAL_TASK_GUIDE_TO_MODEL_TYPES.get(__UpperCamelCase ,set() )
A_ = {
code: name
for code, name in transformers_module.MODEL_NAMES_MAPPING.items()
if (code in model_maping_names or code in special_model_types)
}
return ", ".join([f'''[{name}](../model_doc/{code})''' for code, name in model_names.items()] ) + "\n"
def __snake_case ( __UpperCamelCase : str ,__UpperCamelCase : List[str]=False ):
"""simple docstring"""
A_ , A_ , A_ , A_ = _find_text_in_file(
filename=os.path.join(__UpperCamelCase ,__UpperCamelCase ) ,start_prompt="<!--This tip is automatically generated by `make fix-copies`, do not fill manually!-->" ,end_prompt="<!--End of the generated tip-->" ,)
A_ = get_model_list_for_task(__UpperCamelCase )
if current_list != new_list:
if overwrite:
with open(os.path.join(__UpperCamelCase ,__UpperCamelCase ) ,"w" ,encoding="utf-8" ,newline="\n" ) as f:
f.writelines(lines[:start_index] + [new_list] + lines[end_index:] )
else:
raise ValueError(
f'''The list of models that can be used in the {task_guide} guide needs an update. Run `make fix-copies`'''
" to fix this." )
if __name__ == "__main__":
__a :int = argparse.ArgumentParser()
parser.add_argument('--fix_and_overwrite', action='store_true', help='Whether to fix inconsistencies.')
__a :Optional[Any] = parser.parse_args()
for task_guide in TASK_GUIDE_TO_MODELS.keys():
check_model_list_for_task(task_guide, args.fix_and_overwrite) | 86 | 0 |
import argparse
import json
import math
import os
import time
import traceback
import zipfile
from collections import Counter
import requests
def SCREAMING_SNAKE_CASE__ ( UpperCamelCase__: Dict , UpperCamelCase__: Optional[Any]=None ):
SCREAMING_SNAKE_CASE__ = None
if token is not None:
SCREAMING_SNAKE_CASE__ = {"""Accept""": """application/vnd.github+json""", """Authorization""": f'''Bearer {token}'''}
SCREAMING_SNAKE_CASE__ = f'''https://api.github.com/repos/huggingface/transformers/actions/runs/{workflow_run_id}/jobs?per_page=100'''
SCREAMING_SNAKE_CASE__ = requests.get(UpperCamelCase__ , headers=UpperCamelCase__ ).json()
SCREAMING_SNAKE_CASE__ = {}
try:
job_links.update({job["""name"""]: job["""html_url"""] for job in result["""jobs"""]} )
SCREAMING_SNAKE_CASE__ = math.ceil((result["""total_count"""] - 100) / 100 )
for i in range(UpperCamelCase__ ):
SCREAMING_SNAKE_CASE__ = requests.get(url + f'''&page={i + 2}''' , headers=UpperCamelCase__ ).json()
job_links.update({job["""name"""]: job["""html_url"""] for job in result["""jobs"""]} )
return job_links
except Exception:
print(f'''Unknown error, could not fetch links:\n{traceback.format_exc()}''' )
return {}
def SCREAMING_SNAKE_CASE__ ( UpperCamelCase__: Dict , UpperCamelCase__: str=None ):
SCREAMING_SNAKE_CASE__ = None
if token is not None:
SCREAMING_SNAKE_CASE__ = {"""Accept""": """application/vnd.github+json""", """Authorization""": f'''Bearer {token}'''}
SCREAMING_SNAKE_CASE__ = f'''https://api.github.com/repos/huggingface/transformers/actions/runs/{worflow_run_id}/artifacts?per_page=100'''
SCREAMING_SNAKE_CASE__ = requests.get(UpperCamelCase__ , headers=UpperCamelCase__ ).json()
SCREAMING_SNAKE_CASE__ = {}
try:
artifacts.update({artifact["""name"""]: artifact["""archive_download_url"""] for artifact in result["""artifacts"""]} )
SCREAMING_SNAKE_CASE__ = math.ceil((result["""total_count"""] - 100) / 100 )
for i in range(UpperCamelCase__ ):
SCREAMING_SNAKE_CASE__ = requests.get(url + f'''&page={i + 2}''' , headers=UpperCamelCase__ ).json()
artifacts.update({artifact["""name"""]: artifact["""archive_download_url"""] for artifact in result["""artifacts"""]} )
return artifacts
except Exception:
print(f'''Unknown error, could not fetch links:\n{traceback.format_exc()}''' )
return {}
def SCREAMING_SNAKE_CASE__ ( UpperCamelCase__: Any , UpperCamelCase__: List[Any] , UpperCamelCase__: Tuple , UpperCamelCase__: Optional[Any] ):
SCREAMING_SNAKE_CASE__ = None
if token is not None:
SCREAMING_SNAKE_CASE__ = {"""Accept""": """application/vnd.github+json""", """Authorization""": f'''Bearer {token}'''}
SCREAMING_SNAKE_CASE__ = requests.get(UpperCamelCase__ , headers=UpperCamelCase__ , allow_redirects=UpperCamelCase__ )
SCREAMING_SNAKE_CASE__ = result.headers["""Location"""]
SCREAMING_SNAKE_CASE__ = requests.get(UpperCamelCase__ , allow_redirects=UpperCamelCase__ )
SCREAMING_SNAKE_CASE__ = os.path.join(UpperCamelCase__ , f'''{artifact_name}.zip''' )
with open(UpperCamelCase__ , """wb""" ) as fp:
fp.write(response.content )
def SCREAMING_SNAKE_CASE__ ( UpperCamelCase__: int , UpperCamelCase__: Any=None ):
SCREAMING_SNAKE_CASE__ = []
SCREAMING_SNAKE_CASE__ = []
SCREAMING_SNAKE_CASE__ = None
with zipfile.ZipFile(UpperCamelCase__ ) as z:
for filename in z.namelist():
if not os.path.isdir(UpperCamelCase__ ):
# read the file
if filename in ["failures_line.txt", "summary_short.txt", "job_name.txt"]:
with z.open(UpperCamelCase__ ) as f:
for line in f:
SCREAMING_SNAKE_CASE__ = line.decode("""UTF-8""" ).strip()
if filename == "failures_line.txt":
try:
# `error_line` is the place where `error` occurs
SCREAMING_SNAKE_CASE__ = line[: line.index(""": """ )]
SCREAMING_SNAKE_CASE__ = line[line.index(""": """ ) + len(""": """ ) :]
errors.append([error_line, error] )
except Exception:
# skip un-related lines
pass
elif filename == "summary_short.txt" and line.startswith("""FAILED """ ):
# `test` is the test method that failed
SCREAMING_SNAKE_CASE__ = line[len("""FAILED """ ) :]
failed_tests.append(UpperCamelCase__ )
elif filename == "job_name.txt":
SCREAMING_SNAKE_CASE__ = line
if len(UpperCamelCase__ ) != len(UpperCamelCase__ ):
raise ValueError(
f'''`errors` and `failed_tests` should have the same number of elements. Got {len(UpperCamelCase__ )} for `errors` '''
f'''and {len(UpperCamelCase__ )} for `failed_tests` instead. The test reports in {artifact_zip_path} have some'''
""" problem.""" )
SCREAMING_SNAKE_CASE__ = None
if job_name and job_links:
SCREAMING_SNAKE_CASE__ = job_links.get(UpperCamelCase__ , UpperCamelCase__ )
# A list with elements of the form (line of error, error, failed test)
SCREAMING_SNAKE_CASE__ = [x + [y] + [job_link] for x, y in zip(UpperCamelCase__ , UpperCamelCase__ )]
return result
def SCREAMING_SNAKE_CASE__ ( UpperCamelCase__: Tuple , UpperCamelCase__: Any=None ):
SCREAMING_SNAKE_CASE__ = []
SCREAMING_SNAKE_CASE__ = [os.path.join(UpperCamelCase__ , UpperCamelCase__ ) for p in os.listdir(UpperCamelCase__ ) if p.endswith(""".zip""" )]
for p in paths:
errors.extend(get_errors_from_single_artifact(UpperCamelCase__ , job_links=UpperCamelCase__ ) )
return errors
def SCREAMING_SNAKE_CASE__ ( UpperCamelCase__: int , UpperCamelCase__: str=None ):
SCREAMING_SNAKE_CASE__ = Counter()
counter.update([x[1] for x in logs] )
SCREAMING_SNAKE_CASE__ = counter.most_common()
SCREAMING_SNAKE_CASE__ = {}
for error, count in counts:
if error_filter is None or error not in error_filter:
SCREAMING_SNAKE_CASE__ = {"""count""": count, """failed_tests""": [(x[2], x[0]) for x in logs if x[1] == error]}
SCREAMING_SNAKE_CASE__ = dict(sorted(r.items() , key=lambda UpperCamelCase__ : item[1]["count"] , reverse=UpperCamelCase__ ) )
return r
def SCREAMING_SNAKE_CASE__ ( UpperCamelCase__: Tuple ):
SCREAMING_SNAKE_CASE__ = test.split("""::""" )[0]
if test.startswith("""tests/models/""" ):
SCREAMING_SNAKE_CASE__ = test.split("""/""" )[2]
else:
SCREAMING_SNAKE_CASE__ = None
return test
def SCREAMING_SNAKE_CASE__ ( UpperCamelCase__: Tuple , UpperCamelCase__: Any=None ):
SCREAMING_SNAKE_CASE__ = [(x[0], x[1], get_model(x[2] )) for x in logs]
SCREAMING_SNAKE_CASE__ = [x for x in logs if x[2] is not None]
SCREAMING_SNAKE_CASE__ = {x[2] for x in logs}
SCREAMING_SNAKE_CASE__ = {}
for test in tests:
SCREAMING_SNAKE_CASE__ = Counter()
# count by errors in `test`
counter.update([x[1] for x in logs if x[2] == test] )
SCREAMING_SNAKE_CASE__ = counter.most_common()
SCREAMING_SNAKE_CASE__ = {error: count for error, count in counts if (error_filter is None or error not in error_filter)}
SCREAMING_SNAKE_CASE__ = sum(error_counts.values() )
if n_errors > 0:
SCREAMING_SNAKE_CASE__ = {"""count""": n_errors, """errors""": error_counts}
SCREAMING_SNAKE_CASE__ = dict(sorted(r.items() , key=lambda UpperCamelCase__ : item[1]["count"] , reverse=UpperCamelCase__ ) )
return r
def SCREAMING_SNAKE_CASE__ ( UpperCamelCase__: Tuple ):
SCREAMING_SNAKE_CASE__ = """| no. | error | status |"""
SCREAMING_SNAKE_CASE__ = """|-:|:-|:-|"""
SCREAMING_SNAKE_CASE__ = [header, sep]
for error in reduced_by_error:
SCREAMING_SNAKE_CASE__ = reduced_by_error[error]["""count"""]
SCREAMING_SNAKE_CASE__ = f'''| {count} | {error[:100]} | |'''
lines.append(UpperCamelCase__ )
return "\n".join(UpperCamelCase__ )
def SCREAMING_SNAKE_CASE__ ( UpperCamelCase__: Tuple ):
SCREAMING_SNAKE_CASE__ = """| model | no. of errors | major error | count |"""
SCREAMING_SNAKE_CASE__ = """|-:|-:|-:|-:|"""
SCREAMING_SNAKE_CASE__ = [header, sep]
for model in reduced_by_model:
SCREAMING_SNAKE_CASE__ = reduced_by_model[model]["""count"""]
SCREAMING_SNAKE_CASE__ , SCREAMING_SNAKE_CASE__ = list(reduced_by_model[model]["""errors"""].items() )[0]
SCREAMING_SNAKE_CASE__ = f'''| {model} | {count} | {error[:60]} | {_count} |'''
lines.append(UpperCamelCase__ )
return "\n".join(UpperCamelCase__ )
if __name__ == "__main__":
_lowerCamelCase = argparse.ArgumentParser()
# Required parameters
parser.add_argument('--workflow_run_id', type=str, required=True, help='A GitHub Actions workflow run id.')
parser.add_argument(
'--output_dir',
type=str,
required=True,
help='Where to store the downloaded artifacts and other result files.',
)
parser.add_argument('--token', default=None, type=str, help='A token that has actions:read permission.')
_lowerCamelCase = parser.parse_args()
os.makedirs(args.output_dir, exist_ok=True)
_lowerCamelCase = get_job_links(args.workflow_run_id, token=args.token)
_lowerCamelCase = {}
# To deal with `workflow_call` event, where a job name is the combination of the job names in the caller and callee.
# For example, `PyTorch 1.11 / Model tests (models/albert, single-gpu)`.
if _job_links:
for k, v in _job_links.items():
# This is how GitHub actions combine job names.
if " / " in k:
_lowerCamelCase = k.find(' / ')
_lowerCamelCase = k[index + len(' / ') :]
_lowerCamelCase = v
with open(os.path.join(args.output_dir, 'job_links.json'), 'w', encoding='UTF-8') as fp:
json.dump(job_links, fp, ensure_ascii=False, indent=4)
_lowerCamelCase = get_artifacts_links(args.workflow_run_id, token=args.token)
with open(os.path.join(args.output_dir, 'artifacts.json'), 'w', encoding='UTF-8') as fp:
json.dump(artifacts, fp, ensure_ascii=False, indent=4)
for idx, (name, url) in enumerate(artifacts.items()):
download_artifact(name, url, args.output_dir, args.token)
# Be gentle to GitHub
time.sleep(1)
_lowerCamelCase = get_all_errors(args.output_dir, job_links=job_links)
# `e[1]` is the error
_lowerCamelCase = Counter()
counter.update([e[1] for e in errors])
# print the top 30 most common test errors
_lowerCamelCase = counter.most_common(30)
for item in most_common:
print(item)
with open(os.path.join(args.output_dir, 'errors.json'), 'w', encoding='UTF-8') as fp:
json.dump(errors, fp, ensure_ascii=False, indent=4)
_lowerCamelCase = reduce_by_error(errors)
_lowerCamelCase = reduce_by_model(errors)
_lowerCamelCase = make_github_table(reduced_by_error)
_lowerCamelCase = make_github_table_per_model(reduced_by_model)
with open(os.path.join(args.output_dir, 'reduced_by_error.txt'), 'w', encoding='UTF-8') as fp:
fp.write(sa)
with open(os.path.join(args.output_dir, 'reduced_by_model.txt'), 'w', encoding='UTF-8') as fp:
fp.write(sa) | 6 |
import argparse
import json
from pathlib import Path
import requests
import timm
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from transformers import DeiTImageProcessor, ViTConfig, ViTForImageClassification, ViTImageProcessor, ViTModel
from transformers.utils import logging
logging.set_verbosity_info()
__a :Dict = logging.get_logger(__name__)
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : Tuple=False ):
"""simple docstring"""
A_ = []
for i in range(config.num_hidden_layers ):
# encoder layers: output projection, 2 feedforward neural networks and 2 layernorms
rename_keys.append((f'''blocks.{i}.norm1.weight''', f'''vit.encoder.layer.{i}.layernorm_before.weight''') )
rename_keys.append((f'''blocks.{i}.norm1.bias''', f'''vit.encoder.layer.{i}.layernorm_before.bias''') )
rename_keys.append((f'''blocks.{i}.attn.proj.weight''', f'''vit.encoder.layer.{i}.attention.output.dense.weight''') )
rename_keys.append((f'''blocks.{i}.attn.proj.bias''', f'''vit.encoder.layer.{i}.attention.output.dense.bias''') )
rename_keys.append((f'''blocks.{i}.norm2.weight''', f'''vit.encoder.layer.{i}.layernorm_after.weight''') )
rename_keys.append((f'''blocks.{i}.norm2.bias''', f'''vit.encoder.layer.{i}.layernorm_after.bias''') )
rename_keys.append((f'''blocks.{i}.mlp.fc1.weight''', f'''vit.encoder.layer.{i}.intermediate.dense.weight''') )
rename_keys.append((f'''blocks.{i}.mlp.fc1.bias''', f'''vit.encoder.layer.{i}.intermediate.dense.bias''') )
rename_keys.append((f'''blocks.{i}.mlp.fc2.weight''', f'''vit.encoder.layer.{i}.output.dense.weight''') )
rename_keys.append((f'''blocks.{i}.mlp.fc2.bias''', f'''vit.encoder.layer.{i}.output.dense.bias''') )
# projection layer + position embeddings
rename_keys.extend(
[
("cls_token", "vit.embeddings.cls_token"),
("patch_embed.proj.weight", "vit.embeddings.patch_embeddings.projection.weight"),
("patch_embed.proj.bias", "vit.embeddings.patch_embeddings.projection.bias"),
("pos_embed", "vit.embeddings.position_embeddings"),
] )
if base_model:
# layernorm + pooler
rename_keys.extend(
[
("norm.weight", "layernorm.weight"),
("norm.bias", "layernorm.bias"),
("pre_logits.fc.weight", "pooler.dense.weight"),
("pre_logits.fc.bias", "pooler.dense.bias"),
] )
# if just the base model, we should remove "vit" from all keys that start with "vit"
A_ = [(pair[0], pair[1][4:]) if pair[1].startswith("vit" ) else pair for pair in rename_keys]
else:
# layernorm + classification head
rename_keys.extend(
[
("norm.weight", "vit.layernorm.weight"),
("norm.bias", "vit.layernorm.bias"),
("head.weight", "classifier.weight"),
("head.bias", "classifier.bias"),
] )
return rename_keys
def __snake_case ( __UpperCamelCase : Optional[int] ,__UpperCamelCase : Tuple ,__UpperCamelCase : Any=False ):
"""simple docstring"""
for i in range(config.num_hidden_layers ):
if base_model:
A_ = ""
else:
A_ = "vit."
# read in weights + bias of input projection layer (in timm, this is a single matrix + bias)
A_ = state_dict.pop(f'''blocks.{i}.attn.qkv.weight''' )
A_ = state_dict.pop(f'''blocks.{i}.attn.qkv.bias''' )
# next, add query, keys and values (in that order) to the state dict
A_ = in_proj_weight[
: config.hidden_size, :
]
A_ = in_proj_bias[: config.hidden_size]
A_ = in_proj_weight[
config.hidden_size : config.hidden_size * 2, :
]
A_ = in_proj_bias[
config.hidden_size : config.hidden_size * 2
]
A_ = in_proj_weight[
-config.hidden_size :, :
]
A_ = in_proj_bias[-config.hidden_size :]
def __snake_case ( __UpperCamelCase : List[Any] ):
"""simple docstring"""
A_ = ["head.weight", "head.bias"]
for k in ignore_keys:
state_dict.pop(__UpperCamelCase ,__UpperCamelCase )
def __snake_case ( __UpperCamelCase : Any ,__UpperCamelCase : Optional[int] ,__UpperCamelCase : List[str] ):
"""simple docstring"""
A_ = dct.pop(__UpperCamelCase )
A_ = val
def __snake_case ( ):
"""simple docstring"""
A_ = "http://images.cocodataset.org/val2017/000000039769.jpg"
A_ = Image.open(requests.get(__UpperCamelCase ,stream=__UpperCamelCase ).raw )
return im
@torch.no_grad()
def __snake_case ( __UpperCamelCase : Union[str, Any] ,__UpperCamelCase : Optional[int] ):
"""simple docstring"""
A_ = ViTConfig()
A_ = False
# dataset (ImageNet-21k only or also fine-tuned on ImageNet 2012), patch_size and image_size
if vit_name[-5:] == "in21k":
A_ = True
A_ = int(vit_name[-12:-10] )
A_ = int(vit_name[-9:-6] )
else:
A_ = 1000
A_ = "huggingface/label-files"
A_ = "imagenet-1k-id2label.json"
A_ = json.load(open(hf_hub_download(__UpperCamelCase ,__UpperCamelCase ,repo_type="dataset" ) ,"r" ) )
A_ = {int(__UpperCamelCase ): v for k, v in idalabel.items()}
A_ = idalabel
A_ = {v: k for k, v in idalabel.items()}
A_ = int(vit_name[-6:-4] )
A_ = int(vit_name[-3:] )
# size of the architecture
if "deit" in vit_name:
if vit_name[9:].startswith("tiny" ):
A_ = 192
A_ = 768
A_ = 12
A_ = 3
elif vit_name[9:].startswith("small" ):
A_ = 384
A_ = 1536
A_ = 12
A_ = 6
else:
pass
else:
if vit_name[4:].startswith("small" ):
A_ = 768
A_ = 2304
A_ = 8
A_ = 8
elif vit_name[4:].startswith("base" ):
pass
elif vit_name[4:].startswith("large" ):
A_ = 1024
A_ = 4096
A_ = 24
A_ = 16
elif vit_name[4:].startswith("huge" ):
A_ = 1280
A_ = 5120
A_ = 32
A_ = 16
# load original model from timm
A_ = timm.create_model(__UpperCamelCase ,pretrained=__UpperCamelCase )
timm_model.eval()
# load state_dict of original model, remove and rename some keys
A_ = timm_model.state_dict()
if base_model:
remove_classification_head_(__UpperCamelCase )
A_ = create_rename_keys(__UpperCamelCase ,__UpperCamelCase )
for src, dest in rename_keys:
rename_key(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
read_in_q_k_v(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
# load HuggingFace model
if vit_name[-5:] == "in21k":
A_ = ViTModel(__UpperCamelCase ).eval()
else:
A_ = ViTForImageClassification(__UpperCamelCase ).eval()
model.load_state_dict(__UpperCamelCase )
# Check outputs on an image, prepared by ViTImageProcessor/DeiTImageProcessor
if "deit" in vit_name:
A_ = DeiTImageProcessor(size=config.image_size )
else:
A_ = ViTImageProcessor(size=config.image_size )
A_ = image_processor(images=prepare_img() ,return_tensors="pt" )
A_ = encoding["pixel_values"]
A_ = model(__UpperCamelCase )
if base_model:
A_ = timm_model.forward_features(__UpperCamelCase )
assert timm_pooled_output.shape == outputs.pooler_output.shape
assert torch.allclose(__UpperCamelCase ,outputs.pooler_output ,atol=1E-3 )
else:
A_ = timm_model(__UpperCamelCase )
assert timm_logits.shape == outputs.logits.shape
assert torch.allclose(__UpperCamelCase ,outputs.logits ,atol=1E-3 )
Path(__UpperCamelCase ).mkdir(exist_ok=__UpperCamelCase )
print(f'''Saving model {vit_name} to {pytorch_dump_folder_path}''' )
model.save_pretrained(__UpperCamelCase )
print(f'''Saving image processor to {pytorch_dump_folder_path}''' )
image_processor.save_pretrained(__UpperCamelCase )
if __name__ == "__main__":
__a :str = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
'--vit_name',
default='vit_base_patch16_224',
type=str,
help='Name of the ViT timm model you\'d like to convert.',
)
parser.add_argument(
'--pytorch_dump_folder_path', default=None, type=str, help='Path to the output PyTorch model directory.'
)
__a :Optional[int] = parser.parse_args()
convert_vit_checkpoint(args.vit_name, args.pytorch_dump_folder_path) | 86 | 0 |
"""simple docstring"""
import argparse
import json
from dataclasses import dataclass, field
from functools import partial
from pathlib import Path
from typing import List
import timm
import torch
import torch.nn as nn
from huggingface_hub import hf_hub_download
from torch import Tensor
from transformers import AutoImageProcessor, ResNetConfig, ResNetForImageClassification
from transformers.utils import logging
logging.set_verbosity_info()
a = logging.get_logger()
@dataclass
class lowercase_ :
'''simple docstring'''
UpperCAmelCase : nn.Module
UpperCAmelCase : List[nn.Module] = field(default_factory=__lowerCAmelCase )
UpperCAmelCase : list = field(default_factory=__lowerCAmelCase )
def lowerCAmelCase_ ( self : str , _UpperCAmelCase : Dict , _UpperCAmelCase : Tensor , _UpperCAmelCase : Tensor ):
_A = len(list(m.modules() ) ) == 1 or isinstance(_UpperCAmelCase , nn.Convad ) or isinstance(_UpperCAmelCase , nn.BatchNormad )
if has_not_submodules:
self.traced.append(_UpperCAmelCase )
def __call__( self : List[str] , _UpperCAmelCase : Tensor ):
for m in self.module.modules():
self.handles.append(m.register_forward_hook(self._forward_hook ) )
self.module(_UpperCAmelCase )
[x.remove() for x in self.handles]
return self
@property
def lowerCAmelCase_ ( self : Tuple ):
# check the len of the state_dict keys to see if we have learnable params
return list(filter(lambda _UpperCAmelCase : len(list(x.state_dict().keys() ) ) > 0 , self.traced ) )
@dataclass
class lowercase_ :
'''simple docstring'''
UpperCAmelCase : nn.Module
UpperCAmelCase : nn.Module
UpperCAmelCase : int = 0
UpperCAmelCase : List = field(default_factory=__lowerCAmelCase )
UpperCAmelCase : List = field(default_factory=__lowerCAmelCase )
def __call__( self : Dict , _UpperCAmelCase : Tensor ):
_A = Tracker(self.dest )(_UpperCAmelCase ).parametrized
_A = Tracker(self.src )(_UpperCAmelCase ).parametrized
_A = list(filter(lambda _UpperCAmelCase : type(_UpperCAmelCase ) not in self.src_skip , _UpperCAmelCase ) )
_A = list(filter(lambda _UpperCAmelCase : type(_UpperCAmelCase ) not in self.dest_skip , _UpperCAmelCase ) )
if len(_UpperCAmelCase ) != len(_UpperCAmelCase ):
raise Exception(
F'''Numbers of operations are different. Source module has {len(_UpperCAmelCase )} operations while'''
F''' destination module has {len(_UpperCAmelCase )}.''' )
for dest_m, src_m in zip(_UpperCAmelCase , _UpperCAmelCase ):
dest_m.load_state_dict(src_m.state_dict() )
if self.verbose == 1:
print(F'''Transfered from={src_m} to={dest_m}''' )
def _snake_case ( _snake_case : str , _snake_case : ResNetConfig , _snake_case : Path , _snake_case : bool = True ) -> Union[str, Any]:
'''simple docstring'''
print(F'''Converting {name}...''' )
with torch.no_grad():
_A = timm.create_model(_snake_case , pretrained=_snake_case ).eval()
_A = ResNetForImageClassification(_snake_case ).eval()
_A = ModuleTransfer(src=_snake_case , dest=_snake_case )
_A = torch.randn((1, 3, 2_24, 2_24) )
module_transfer(_snake_case )
assert torch.allclose(from_model(_snake_case ) , our_model(_snake_case ).logits ), "The model logits don't match the original one."
_A = F'''resnet{"-".join(name.split("resnet" ) )}'''
print(_snake_case )
if push_to_hub:
our_model.push_to_hub(
repo_path_or_name=save_directory / checkpoint_name , commit_message='Add model' , use_temp_dir=_snake_case , )
# we can use the convnext one
_A = AutoImageProcessor.from_pretrained('facebook/convnext-base-224-22k-1k' )
image_processor.push_to_hub(
repo_path_or_name=save_directory / checkpoint_name , commit_message='Add image processor' , use_temp_dir=_snake_case , )
print(F'''Pushed {checkpoint_name}''' )
def _snake_case ( _snake_case : Path , _snake_case : str = None , _snake_case : bool = True ) -> Tuple:
'''simple docstring'''
_A = 'imagenet-1k-id2label.json'
_A = 10_00
_A = (1, num_labels)
_A = 'huggingface/label-files'
_A = num_labels
_A = json.load(open(hf_hub_download(_snake_case , _snake_case , repo_type='dataset' ) , 'r' ) )
_A = {int(_snake_case ): v for k, v in idalabel.items()}
_A = idalabel
_A = {v: k for k, v in idalabel.items()}
_A = partial(_snake_case , num_labels=_snake_case , idalabel=_snake_case , labelaid=_snake_case )
_A = {
'resnet18': ImageNetPreTrainedConfig(
depths=[2, 2, 2, 2] , hidden_sizes=[64, 1_28, 2_56, 5_12] , layer_type='basic' ),
'resnet26': ImageNetPreTrainedConfig(
depths=[2, 2, 2, 2] , hidden_sizes=[2_56, 5_12, 10_24, 20_48] , layer_type='bottleneck' ),
'resnet34': ImageNetPreTrainedConfig(
depths=[3, 4, 6, 3] , hidden_sizes=[64, 1_28, 2_56, 5_12] , layer_type='basic' ),
'resnet50': ImageNetPreTrainedConfig(
depths=[3, 4, 6, 3] , hidden_sizes=[2_56, 5_12, 10_24, 20_48] , layer_type='bottleneck' ),
'resnet101': ImageNetPreTrainedConfig(
depths=[3, 4, 23, 3] , hidden_sizes=[2_56, 5_12, 10_24, 20_48] , layer_type='bottleneck' ),
'resnet152': ImageNetPreTrainedConfig(
depths=[3, 8, 36, 3] , hidden_sizes=[2_56, 5_12, 10_24, 20_48] , layer_type='bottleneck' ),
}
if model_name:
convert_weight_and_push(_snake_case , names_to_config[model_name] , _snake_case , _snake_case )
else:
for model_name, config in names_to_config.items():
convert_weight_and_push(_snake_case , _snake_case , _snake_case , _snake_case )
return config, expected_shape
if __name__ == "__main__":
a = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
'''--model_name''',
default=None,
type=str,
help=(
'''The name of the model you wish to convert, it must be one of the supported resnet* architecture,'''
''' currently: resnet18,26,34,50,101,152. If `None`, all of them will the converted.'''
),
)
parser.add_argument(
'''--pytorch_dump_folder_path''',
default=None,
type=Path,
required=True,
help='''Path to the output PyTorch model directory.''',
)
parser.add_argument(
'''--push_to_hub''',
default=True,
type=bool,
required=False,
help='''If True, push model and image processor to the hub.''',
)
a = parser.parse_args()
a = args.pytorch_dump_folder_path
pytorch_dump_folder_path.mkdir(exist_ok=True, parents=True)
convert_weights_and_push(pytorch_dump_folder_path, args.model_name, args.push_to_hub)
| 7 |
def __snake_case ( __UpperCamelCase : int = 50 ):
"""simple docstring"""
A_ = [1] * (length + 1)
for row_length in range(length + 1 ):
for tile_length in range(2 ,5 ):
for tile_start in range(row_length - tile_length + 1 ):
ways_number[row_length] += ways_number[
row_length - tile_start - tile_length
]
return ways_number[length]
if __name__ == "__main__":
print(F"{solution() = }") | 86 | 0 |
'''simple docstring'''
import random
def _lowerCAmelCase ( __snake_case : int ) -> bool:
__A : Tuple = num - 1
__A : Optional[Any] = 0
while s % 2 == 0:
__A : Optional[int] = s // 2
t += 1
for _ in range(5 ):
__A : List[str] = random.randrange(2 , num - 1 )
__A : str = pow(__snake_case , __snake_case , __snake_case )
if v != 1:
__A : Optional[int] = 0
while v != (num - 1):
if i == t - 1:
return False
else:
__A : Optional[int] = i + 1
__A : Optional[int] = (v**2) % num
return True
def _lowerCAmelCase ( __snake_case : int ) -> bool:
if num < 2:
return False
__A : Optional[int] = [
2,
3,
5,
7,
11,
13,
17,
19,
23,
29,
31,
37,
41,
43,
47,
53,
59,
61,
67,
71,
73,
79,
83,
89,
97,
1_01,
1_03,
1_07,
1_09,
1_13,
1_27,
1_31,
1_37,
1_39,
1_49,
1_51,
1_57,
1_63,
1_67,
1_73,
1_79,
1_81,
1_91,
1_93,
1_97,
1_99,
2_11,
2_23,
2_27,
2_29,
2_33,
2_39,
2_41,
2_51,
2_57,
2_63,
2_69,
2_71,
2_77,
2_81,
2_83,
2_93,
3_07,
3_11,
3_13,
3_17,
3_31,
3_37,
3_47,
3_49,
3_53,
3_59,
3_67,
3_73,
3_79,
3_83,
3_89,
3_97,
4_01,
4_09,
4_19,
4_21,
4_31,
4_33,
4_39,
4_43,
4_49,
4_57,
4_61,
4_63,
4_67,
4_79,
4_87,
4_91,
4_99,
5_03,
5_09,
5_21,
5_23,
5_41,
5_47,
5_57,
5_63,
5_69,
5_71,
5_77,
5_87,
5_93,
5_99,
6_01,
6_07,
6_13,
6_17,
6_19,
6_31,
6_41,
6_43,
6_47,
6_53,
6_59,
6_61,
6_73,
6_77,
6_83,
6_91,
7_01,
7_09,
7_19,
7_27,
7_33,
7_39,
7_43,
7_51,
7_57,
7_61,
7_69,
7_73,
7_87,
7_97,
8_09,
8_11,
8_21,
8_23,
8_27,
8_29,
8_39,
8_53,
8_57,
8_59,
8_63,
8_77,
8_81,
8_83,
8_87,
9_07,
9_11,
9_19,
9_29,
9_37,
9_41,
9_47,
9_53,
9_67,
9_71,
9_77,
9_83,
9_91,
9_97,
]
if num in low_primes:
return True
for prime in low_primes:
if (num % prime) == 0:
return False
return rabin_miller(__snake_case )
def _lowerCAmelCase ( __snake_case : int = 10_24 ) -> int:
while True:
__A : Any = random.randrange(2 ** (keysize - 1) , 2 ** (keysize) )
if is_prime_low_num(__snake_case ):
return num
if __name__ == "__main__":
lowercase__ : List[str] = generate_large_prime()
print(('''Prime number:''', num))
print(('''is_prime_low_num:''', is_prime_low_num(num))) | 8 |
from typing import Any, Dict, List, Union
from ..utils import add_end_docstrings, is_torch_available, is_vision_available, logging, requires_backends
from .base import PIPELINE_INIT_ARGS, ChunkPipeline
if is_vision_available():
from PIL import Image
from ..image_utils import load_image
if is_torch_available():
import torch
from transformers.modeling_outputs import BaseModelOutput
from ..models.auto.modeling_auto import MODEL_FOR_ZERO_SHOT_OBJECT_DETECTION_MAPPING
__a :List[str] = logging.get_logger(__name__)
@add_end_docstrings(snake_case_ )
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Any , **UpperCAmelCase : List[str] ):
super().__init__(**UpperCAmelCase )
if self.framework == "tf":
raise ValueError(f'''The {self.__class__} is only available in PyTorch.''' )
requires_backends(self , "vision" )
self.check_model_type(UpperCAmelCase )
def __call__( self : Optional[int] , UpperCAmelCase : Union[str, "Image.Image", List[Dict[str, Any]]] , UpperCAmelCase : Union[str, List[str]] = None , **UpperCAmelCase : List[Any] , ):
if "text_queries" in kwargs:
A_ = kwargs.pop("text_queries" )
if isinstance(UpperCAmelCase , (str, Image.Image) ):
A_ = {"image": image, "candidate_labels": candidate_labels}
else:
A_ = image
A_ = super().__call__(UpperCAmelCase , **UpperCAmelCase )
return results
def __A ( self : int , **UpperCAmelCase : Tuple ):
A_ = {}
if "threshold" in kwargs:
A_ = kwargs["threshold"]
if "top_k" in kwargs:
A_ = kwargs["top_k"]
return {}, {}, postprocess_params
def __A ( self : List[str] , UpperCAmelCase : Dict ):
A_ = load_image(inputs["image"] )
A_ = inputs["candidate_labels"]
if isinstance(UpperCAmelCase , UpperCAmelCase ):
A_ = candidate_labels.split("," )
A_ = torch.tensor([[image.height, image.width]] , dtype=torch.intaa )
for i, candidate_label in enumerate(UpperCAmelCase ):
A_ = self.tokenizer(UpperCAmelCase , return_tensors=self.framework )
A_ = self.image_processor(UpperCAmelCase , return_tensors=self.framework )
yield {
"is_last": i == len(UpperCAmelCase ) - 1,
"target_size": target_size,
"candidate_label": candidate_label,
**text_inputs,
**image_features,
}
def __A ( self : str , UpperCAmelCase : int ):
A_ = model_inputs.pop("target_size" )
A_ = model_inputs.pop("candidate_label" )
A_ = model_inputs.pop("is_last" )
A_ = self.model(**UpperCAmelCase )
A_ = {"target_size": target_size, "candidate_label": candidate_label, "is_last": is_last, **outputs}
return model_outputs
def __A ( self : Dict , UpperCAmelCase : Any , UpperCAmelCase : Optional[int]=0.1 , UpperCAmelCase : Optional[int]=None ):
A_ = []
for model_output in model_outputs:
A_ = model_output["candidate_label"]
A_ = BaseModelOutput(UpperCAmelCase )
A_ = self.image_processor.post_process_object_detection(
outputs=UpperCAmelCase , threshold=UpperCAmelCase , target_sizes=model_output["target_size"] )[0]
for index in outputs["scores"].nonzero():
A_ = outputs["scores"][index].item()
A_ = self._get_bounding_box(outputs["boxes"][index][0] )
A_ = {"score": score, "label": label, "box": box}
results.append(UpperCAmelCase )
A_ = sorted(UpperCAmelCase , key=lambda UpperCAmelCase : x["score"] , reverse=UpperCAmelCase )
if top_k:
A_ = results[:top_k]
return results
def __A ( self : List[str] , UpperCAmelCase : "torch.Tensor" ):
if self.framework != "pt":
raise ValueError("The ZeroShotObjectDetectionPipeline is only available in PyTorch." )
A_ , A_ , A_ , A_ = box.int().tolist()
A_ = {
"xmin": xmin,
"ymin": ymin,
"xmax": xmax,
"ymax": ymax,
}
return bbox | 86 | 0 |
from typing import TYPE_CHECKING
from ...utils import (
OptionalDependencyNotAvailable,
_LazyModule,
is_sentencepiece_available,
is_tokenizers_available,
is_torch_available,
is_vision_available,
)
SCREAMING_SNAKE_CASE__ = {'''processing_layoutxlm''': ['''LayoutXLMProcessor''']}
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
SCREAMING_SNAKE_CASE__ = ['''LayoutXLMTokenizer''']
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
SCREAMING_SNAKE_CASE__ = ['''LayoutXLMTokenizerFast''']
if TYPE_CHECKING:
from .processing_layoutxlm import LayoutXLMProcessor
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_layoutxlm import LayoutXLMTokenizer
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_layoutxlm_fast import LayoutXLMTokenizerFast
else:
import sys
SCREAMING_SNAKE_CASE__ = _LazyModule(__name__, globals()['''__file__'''], _import_structure, module_spec=__spec__)
| 9 |
import argparse
import torch
# Step 1. clone https://github.com/microsoft/unilm
# Step 2. git checkout to https://github.com/microsoft/unilm/commit/b94ec76c36f02fb2b0bf0dcb0b8554a2185173cd
# Step 3. cd unilm
# Step 4. ln -s $(realpath wavlm/modules.py) ./ # create simlink
# import classes
from unilm.wavlm.WavLM import WavLM as WavLMOrig
from unilm.wavlm.WavLM import WavLMConfig as WavLMConfigOrig
from transformers import WavLMConfig, WavLMModel, logging
logging.set_verbosity_info()
__a :Any = logging.get_logger(__name__)
__a :int = {
'post_extract_proj': 'feature_projection.projection',
'encoder.pos_conv.0': 'encoder.pos_conv_embed.conv',
'self_attn.k_proj': 'encoder.layers.*.attention.k_proj',
'self_attn.v_proj': 'encoder.layers.*.attention.v_proj',
'self_attn.q_proj': 'encoder.layers.*.attention.q_proj',
'self_attn.out_proj': 'encoder.layers.*.attention.out_proj',
'self_attn.grep_linear': 'encoder.layers.*.attention.gru_rel_pos_linear',
'self_attn.relative_attention_bias': 'encoder.layers.*.attention.rel_attn_embed',
'self_attn.grep_a': 'encoder.layers.*.attention.gru_rel_pos_const',
'self_attn_layer_norm': 'encoder.layers.*.layer_norm',
'fc1': 'encoder.layers.*.feed_forward.intermediate_dense',
'fc2': 'encoder.layers.*.feed_forward.output_dense',
'final_layer_norm': 'encoder.layers.*.final_layer_norm',
'encoder.layer_norm': 'encoder.layer_norm',
'w2v_model.layer_norm': 'feature_projection.layer_norm',
'quantizer.weight_proj': 'quantizer.weight_proj',
'quantizer.vars': 'quantizer.codevectors',
'project_q': 'project_q',
'final_proj': 'project_hid',
'w2v_encoder.proj': 'ctc_proj',
'mask_emb': 'masked_spec_embed',
}
__a :Tuple = [
'ctc_proj',
'quantizer.weight_proj',
'quantizer.codevectors',
'project_q',
'project_hid',
]
def __snake_case ( __UpperCamelCase : List[str] ,__UpperCamelCase : List[Any] ,__UpperCamelCase : Dict ,__UpperCamelCase : Any ,__UpperCamelCase : Optional[int] ):
"""simple docstring"""
for attribute in key.split("." ):
A_ = getattr(__UpperCamelCase ,__UpperCamelCase )
if weight_type is not None:
A_ = getattr(__UpperCamelCase ,__UpperCamelCase ).shape
else:
A_ = hf_pointer.shape
assert hf_shape == value.shape, (
f'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
f''' {value.shape} for {full_name}'''
)
if weight_type == "weight":
A_ = value
elif weight_type == "weight_g":
A_ = value
elif weight_type == "weight_v":
A_ = value
elif weight_type == "bias":
A_ = value
else:
A_ = value
logger.info(f'''{key + "." + weight_type if weight_type is not None else ""} was initialized from {full_name}.''' )
def __snake_case ( __UpperCamelCase : Optional[int] ,__UpperCamelCase : Optional[Any] ):
"""simple docstring"""
A_ = []
A_ = fairseq_model.state_dict()
A_ = hf_model.feature_extractor
for name, value in fairseq_dict.items():
A_ = False
if "conv_layers" in name:
load_conv_layer(
__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,hf_model.config.feat_extract_norm == "group" ,)
A_ = True
else:
for key, mapped_key in MAPPING.items():
if key in name or key.split("w2v_model." )[-1] == name.split("." )[0]:
A_ = True
if "*" in mapped_key:
A_ = name.split(__UpperCamelCase )[0].split("." )[-2]
A_ = mapped_key.replace("*" ,__UpperCamelCase )
if "weight_g" in name:
A_ = "weight_g"
elif "weight_v" in name:
A_ = "weight_v"
elif "bias" in name and "relative_attention_bias" not in name:
A_ = "bias"
elif "weight" in name:
# TODO: don't match quantizer.weight_proj
A_ = "weight"
else:
A_ = None
set_recursively(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
continue
if not is_used:
unused_weights.append(__UpperCamelCase )
logger.warning(f'''Unused weights: {unused_weights}''' )
def __snake_case ( __UpperCamelCase : Optional[int] ,__UpperCamelCase : Dict ,__UpperCamelCase : List[str] ,__UpperCamelCase : Union[str, Any] ,__UpperCamelCase : Optional[int] ):
"""simple docstring"""
A_ = full_name.split("conv_layers." )[-1]
A_ = name.split("." )
A_ = int(items[0] )
A_ = int(items[1] )
if type_id == 0:
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.bias.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape, (
f'''{full_name} has size {value.shape}, but {feature_extractor[layer_id].layer_norm.bias.data.shape} was'''
" found."
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor[layer_id].layer_norm.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(__UpperCamelCase )
@torch.no_grad()
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : str ,__UpperCamelCase : int=None ):
"""simple docstring"""
A_ = torch.load(__UpperCamelCase )
A_ = WavLMConfigOrig(checkpoint["cfg"] )
A_ = WavLMOrig(__UpperCamelCase )
model.load_state_dict(checkpoint["model"] )
model.eval()
if config_path is not None:
A_ = WavLMConfig.from_pretrained(__UpperCamelCase )
else:
A_ = WavLMConfig()
A_ = WavLMModel(__UpperCamelCase )
recursively_load_weights(__UpperCamelCase ,__UpperCamelCase )
hf_wavlm.save_pretrained(__UpperCamelCase )
if __name__ == "__main__":
__a :List[Any] = argparse.ArgumentParser()
parser.add_argument('--pytorch_dump_folder_path', default=None, type=str, help='Path to the output PyTorch model.')
parser.add_argument('--checkpoint_path', default=None, type=str, help='Path to fairseq checkpoint')
parser.add_argument('--config_path', default=None, type=str, help='Path to hf config.json of model to convert')
__a :Optional[int] = parser.parse_args()
convert_wavlm_checkpoint(args.checkpoint_path, args.pytorch_dump_folder_path, args.config_path) | 86 | 0 |
import random
def _snake_case ( __snake_case , __snake_case , __snake_case = False ):
_UpperCamelCase = {i: [] for i in range(__snake_case )}
# if probability is greater or equal than 1, then generate a complete graph
if probability >= 1:
return complete_graph(__snake_case )
# if probability is lower or equal than 0, then return a graph without edges
if probability <= 0:
return graph
# for each couple of nodes, add an edge from u to v
# if the number randomly generated is greater than probability probability
for i in range(__snake_case ):
for j in range(i + 1 , __snake_case ):
if random.random() < probability:
graph[i].append(__snake_case )
if not directed:
# if the graph is undirected, add an edge in from j to i, either
graph[j].append(__snake_case )
return graph
def _snake_case ( __snake_case ):
return {
i: [j for j in range(__snake_case ) if i != j] for i in range(__snake_case )
}
if __name__ == "__main__":
import doctest
doctest.testmod()
| 10 |
def __snake_case ( __UpperCamelCase : list ,__UpperCamelCase : int = 0 ):
"""simple docstring"""
A_ = length or len(__UpperCamelCase )
A_ = False
for i in range(length - 1 ):
if list_data[i] > list_data[i + 1]:
A_ , A_ = list_data[i + 1], list_data[i]
A_ = True
return list_data if not swapped else bubble_sort(__UpperCamelCase ,length - 1 )
if __name__ == "__main__":
import doctest
doctest.testmod() | 86 | 0 |
'''simple docstring'''
import collections
import os
import re
from pathlib import Path
lowercase_ = "src/transformers"
# Matches is_xxx_available()
lowercase_ = re.compile(R"is\_([a-z_]*)_available()")
# Catches a one-line _import_struct = {xxx}
lowercase_ = re.compile(R"^_import_structure\s+=\s+\{([^\}]+)\}")
# Catches a line with a key-values pattern: "bla": ["foo", "bar"]
lowercase_ = re.compile(R"\s+\"\S*\":\s+\[([^\]]*)\]")
# Catches a line if not is_foo_available
lowercase_ = re.compile(R"^\s*if\s+not\s+is\_[a-z_]*\_available\(\)")
# Catches a line _import_struct["bla"].append("foo")
lowercase_ = re.compile(R"^\s*_import_structure\[\"\S*\"\]\.append\(\"(\S*)\"\)")
# Catches a line _import_struct["bla"].extend(["foo", "bar"]) or _import_struct["bla"] = ["foo", "bar"]
lowercase_ = re.compile(R"^\s*_import_structure\[\S*\](?:\.extend\(|\s*=\s+)\[([^\]]*)\]")
# Catches a line with an object between quotes and a comma: "MyModel",
lowercase_ = re.compile(R"^\s+\"([^\"]+)\",")
# Catches a line with objects between brackets only: ["foo", "bar"],
lowercase_ = re.compile(R"^\s+\[([^\]]+)\]")
# Catches a line with from foo import bar, bla, boo
lowercase_ = re.compile(R"\s+from\s+\S*\s+import\s+([^\(\s].*)\n")
# Catches a line with try:
lowercase_ = re.compile(R"^\s*try:")
# Catches a line with else:
lowercase_ = re.compile(R"^\s*else:")
def lowerCAmelCase (__A):
"""simple docstring"""
if _re_test_backend.search(__A) is None:
return None
_a = [b[0] for b in _re_backend.findall(__A)]
backends.sort()
return "_and_".join(__A)
def lowerCAmelCase (__A):
"""simple docstring"""
with open(__A , '''r''' , encoding='''utf-8''' , newline='''\n''') as f:
_a = f.readlines()
_a = 0
while line_index < len(__A) and not lines[line_index].startswith('''_import_structure = {'''):
line_index += 1
# If this is a traditional init, just return.
if line_index >= len(__A):
return None
# First grab the objects without a specific backend in _import_structure
_a = []
while not lines[line_index].startswith('''if TYPE_CHECKING''') and find_backend(lines[line_index]) is None:
_a = lines[line_index]
# If we have everything on a single line, let's deal with it.
if _re_one_line_import_struct.search(__A):
_a = _re_one_line_import_struct.search(__A).groups()[0]
_a = re.findall(r'''\[([^\]]+)\]''' , __A)
for imp in imports:
objects.extend([obj[1:-1] for obj in imp.split(''', ''')])
line_index += 1
continue
_a = _re_import_struct_key_value.search(__A)
if single_line_import_search is not None:
_a = [obj[1:-1] for obj in single_line_import_search.groups()[0].split(''', ''') if len(__A) > 0]
objects.extend(__A)
elif line.startswith(''' ''' * 8 + '''"'''):
objects.append(line[9:-3])
line_index += 1
_a = {'''none''': objects}
# Let's continue with backend-specific objects in _import_structure
while not lines[line_index].startswith('''if TYPE_CHECKING'''):
# If the line is an if not is_backend_available, we grab all objects associated.
_a = find_backend(lines[line_index])
# Check if the backend declaration is inside a try block:
if _re_try.search(lines[line_index - 1]) is None:
_a = None
if backend is not None:
line_index += 1
# Scroll until we hit the else block of try-except-else
while _re_else.search(lines[line_index]) is None:
line_index += 1
line_index += 1
_a = []
# Until we unindent, add backend objects to the list
while len(lines[line_index]) <= 1 or lines[line_index].startswith(''' ''' * 4):
_a = lines[line_index]
if _re_import_struct_add_one.search(__A) is not None:
objects.append(_re_import_struct_add_one.search(__A).groups()[0])
elif _re_import_struct_add_many.search(__A) is not None:
_a = _re_import_struct_add_many.search(__A).groups()[0].split(''', ''')
_a = [obj[1:-1] for obj in imports if len(__A) > 0]
objects.extend(__A)
elif _re_between_brackets.search(__A) is not None:
_a = _re_between_brackets.search(__A).groups()[0].split(''', ''')
_a = [obj[1:-1] for obj in imports if len(__A) > 0]
objects.extend(__A)
elif _re_quote_object.search(__A) is not None:
objects.append(_re_quote_object.search(__A).groups()[0])
elif line.startswith(''' ''' * 8 + '''"'''):
objects.append(line[9:-3])
elif line.startswith(''' ''' * 12 + '''"'''):
objects.append(line[13:-3])
line_index += 1
_a = objects
else:
line_index += 1
# At this stage we are in the TYPE_CHECKING part, first grab the objects without a specific backend
_a = []
while (
line_index < len(__A)
and find_backend(lines[line_index]) is None
and not lines[line_index].startswith('''else''')
):
_a = lines[line_index]
_a = _re_import.search(__A)
if single_line_import_search is not None:
objects.extend(single_line_import_search.groups()[0].split(''', '''))
elif line.startswith(''' ''' * 8):
objects.append(line[8:-2])
line_index += 1
_a = {'''none''': objects}
# Let's continue with backend-specific objects
while line_index < len(__A):
# If the line is an if is_backend_available, we grab all objects associated.
_a = find_backend(lines[line_index])
# Check if the backend declaration is inside a try block:
if _re_try.search(lines[line_index - 1]) is None:
_a = None
if backend is not None:
line_index += 1
# Scroll until we hit the else block of try-except-else
while _re_else.search(lines[line_index]) is None:
line_index += 1
line_index += 1
_a = []
# Until we unindent, add backend objects to the list
while len(lines[line_index]) <= 1 or lines[line_index].startswith(''' ''' * 8):
_a = lines[line_index]
_a = _re_import.search(__A)
if single_line_import_search is not None:
objects.extend(single_line_import_search.groups()[0].split(''', '''))
elif line.startswith(''' ''' * 12):
objects.append(line[12:-2])
line_index += 1
_a = objects
else:
line_index += 1
return import_dict_objects, type_hint_objects
def lowerCAmelCase (__A , __A):
"""simple docstring"""
def find_duplicates(__A):
return [k for k, v in collections.Counter(__A).items() if v > 1]
if list(import_dict_objects.keys()) != list(type_hint_objects.keys()):
return ["Both sides of the init do not have the same backends!"]
_a = []
for key in import_dict_objects.keys():
_a = find_duplicates(import_dict_objects[key])
if duplicate_imports:
errors.append(F'''Duplicate _import_structure definitions for: {duplicate_imports}''')
_a = find_duplicates(type_hint_objects[key])
if duplicate_type_hints:
errors.append(F'''Duplicate TYPE_CHECKING objects for: {duplicate_type_hints}''')
if sorted(set(import_dict_objects[key])) != sorted(set(type_hint_objects[key])):
_a = '''base imports''' if key == '''none''' else F'''{key} backend'''
errors.append(F'''Differences for {name}:''')
for a in type_hint_objects[key]:
if a not in import_dict_objects[key]:
errors.append(F''' {a} in TYPE_HINT but not in _import_structure.''')
for a in import_dict_objects[key]:
if a not in type_hint_objects[key]:
errors.append(F''' {a} in _import_structure but not in TYPE_HINT.''')
return errors
def lowerCAmelCase ():
"""simple docstring"""
_a = []
for root, _, files in os.walk(__A):
if "__init__.py" in files:
_a = os.path.join(__A , '''__init__.py''')
_a = parse_init(__A)
if objects is not None:
_a = analyze_results(*__A)
if len(__A) > 0:
_a = F'''Problem in {fname}, both halves do not define the same objects.\n{errors[0]}'''
failures.append('''\n'''.join(__A))
if len(__A) > 0:
raise ValueError('''\n\n'''.join(__A))
def lowerCAmelCase ():
"""simple docstring"""
_a = []
for path, directories, files in os.walk(__A):
for folder in directories:
# Ignore private modules
if folder.startswith('''_'''):
directories.remove(__A)
continue
# Ignore leftovers from branches (empty folders apart from pycache)
if len(list((Path(__A) / folder).glob('''*.py'''))) == 0:
continue
_a = str((Path(__A) / folder).relative_to(__A))
_a = short_path.replace(os.path.sep , '''.''')
submodules.append(__A)
for fname in files:
if fname == "__init__.py":
continue
_a = str((Path(__A) / fname).relative_to(__A))
_a = short_path.replace('''.py''' , '''''').replace(os.path.sep , '''.''')
if len(submodule.split('''.''')) == 1:
submodules.append(__A)
return submodules
lowercase_ = [
"convert_pytorch_checkpoint_to_tf2",
"modeling_flax_pytorch_utils",
"models.esm.openfold_utils",
]
def lowerCAmelCase ():
"""simple docstring"""
from transformers.utils import direct_transformers_import
_a = direct_transformers_import(__A)
_a = set(transformers._import_structure.keys())
# This contains all the base keys of the _import_structure object defined in the init, but if the user is missing
# some optional dependencies, they may not have all of them. Thus we read the init to read all additions and
# (potentiall re-) add them.
with open(os.path.join(__A , '''__init__.py''') , '''r''') as f:
_a = f.read()
import_structure_keys.update(set(re.findall(r'''import_structure\[\"([^\"]*)\"\]''' , __A)))
_a = [
module
for module in get_transformers_submodules()
if module not in IGNORE_SUBMODULES and module not in import_structure_keys
]
if len(__A) > 0:
_a = '''\n'''.join(F'''- {module}''' for module in module_not_registered)
raise ValueError(
'''The following submodules are not properly registed in the main init of Transformers:\n'''
F'''{list_of_modules}\n'''
'''Make sure they appear somewhere in the keys of `_import_structure` with an empty list as value.''')
if __name__ == "__main__":
check_all_inits()
check_submodules()
| 11 |
import pickle
import unittest
import torch
from accelerate import Accelerator
from accelerate.state import AcceleratorState
from accelerate.test_utils import require_cpu
@require_cpu
class _a ( unittest.TestCase ):
"""simple docstring"""
def __A ( self : List[str] ):
A_ = torch.nn.Linear(10 , 10 )
A_ = torch.optim.SGD(model.parameters() , 0.1 )
A_ = Accelerator()
A_ = accelerator.prepare(UpperCAmelCase )
try:
pickle.loads(pickle.dumps(UpperCAmelCase ) )
except Exception as e:
self.fail(f'''Accelerated optimizer pickling failed with {e}''' )
AcceleratorState._reset_state() | 86 | 0 |
def UpperCamelCase ( lowercase_ ) -> int:
'''simple docstring'''
if a < 0:
raise ValueError("""Input value must be a positive integer""" )
elif isinstance(lowercase_ , lowercase_ ):
raise TypeError("""Input value must be a 'int' type""" )
return bin(lowercase_ ).count("""1""" )
if __name__ == "__main__":
import doctest
doctest.testmod()
| 12 |
import argparse
import json
import os
import fairseq
import torch
from torch import nn
from transformers import (
SpeechaTextaConfig,
SpeechaTextaForCausalLM,
SpeechaTextaTokenizer,
SpeechEncoderDecoderConfig,
SpeechEncoderDecoderModel,
WavaVecaConfig,
WavaVecaFeatureExtractor,
WavaVecaModel,
logging,
)
logging.set_verbosity_info()
__a :List[str] = logging.get_logger(__name__)
__a :Optional[int] = {
'post_extract_proj': 'feature_projection.projection',
'encoder.pos_conv.0': 'encoder.pos_conv_embed.conv',
'self_attn.k_proj': 'encoder.layers.*.attention.k_proj',
'self_attn.v_proj': 'encoder.layers.*.attention.v_proj',
'self_attn.q_proj': 'encoder.layers.*.attention.q_proj',
'self_attn.out_proj': 'encoder.layers.*.attention.out_proj',
'self_attn_layer_norm': 'encoder.layers.*.layer_norm',
'fc1': 'encoder.layers.*.feed_forward.intermediate_dense',
'fc2': 'encoder.layers.*.feed_forward.output_dense',
'final_layer_norm': 'encoder.layers.*.final_layer_norm',
'encoder.layer_norm': 'encoder.layer_norm',
'w2v_model.layer_norm': 'feature_projection.layer_norm',
'quantizer.weight_proj': 'quantizer.weight_proj',
'quantizer.vars': 'quantizer.codevectors',
'project_q': 'project_q',
'final_proj': 'project_hid',
'w2v_encoder.proj': 'lm_head',
'mask_emb': 'masked_spec_embed',
}
__a :Any = [
'lm_head',
'quantizer.weight_proj',
'quantizer.codevectors',
'project_q',
'project_hid',
]
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : Optional[Any] ,__UpperCamelCase : Optional[int] ,__UpperCamelCase : List[str] ,__UpperCamelCase : Union[str, Any] ):
"""simple docstring"""
for attribute in key.split("." ):
A_ = getattr(__UpperCamelCase ,__UpperCamelCase )
if weight_type is not None:
A_ = getattr(__UpperCamelCase ,__UpperCamelCase ).shape
else:
A_ = hf_pointer.shape
assert hf_shape == value.shape, (
f'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
f''' {value.shape} for {full_name}'''
)
if weight_type == "weight":
A_ = value
elif weight_type == "weight_g":
A_ = value
elif weight_type == "weight_v":
A_ = value
elif weight_type == "bias":
A_ = value
else:
A_ = value
logger.info(f'''{key + "." + weight_type if weight_type is not None else ""} was initialized from {full_name}.''' )
def __snake_case ( __UpperCamelCase : Union[str, Any] ,__UpperCamelCase : Dict ):
"""simple docstring"""
A_ = []
A_ = fairseq_model.state_dict()
A_ = hf_model.feature_extractor
# if encoder has different dim to decoder -> use proj_weight
A_ = None
for name, value in fairseq_dict.items():
A_ = False
if "conv_layers" in name:
load_conv_layer(
__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,hf_model.config.feat_extract_norm == "group" ,)
A_ = True
elif name.split("." )[0] == "proj":
A_ = fairseq_model.proj
A_ = True
else:
for key, mapped_key in MAPPING.items():
if key in name or key.split("w2v_model." )[-1] == name.split("." )[0]:
A_ = True
if "*" in mapped_key:
A_ = name.split(__UpperCamelCase )[0].split("." )[-2]
A_ = mapped_key.replace("*" ,__UpperCamelCase )
if "weight_g" in name:
A_ = "weight_g"
elif "weight_v" in name:
A_ = "weight_v"
elif "bias" in name:
A_ = "bias"
elif "weight" in name:
A_ = "weight"
else:
A_ = None
set_recursively(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
continue
if not is_used:
unused_weights.append(__UpperCamelCase )
logger.warning(f'''Unused weights: {unused_weights}''' )
return proj_weight
def __snake_case ( __UpperCamelCase : str ,__UpperCamelCase : Optional[Any] ,__UpperCamelCase : int ,__UpperCamelCase : int ,__UpperCamelCase : Any ):
"""simple docstring"""
A_ = full_name.split("conv_layers." )[-1]
A_ = name.split("." )
A_ = int(items[0] )
A_ = int(items[1] )
if type_id == 0:
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.bias.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape, (
f'''{full_name} has size {value.shape}, but {feature_extractor[layer_id].layer_norm.bias.data.shape} was'''
" found."
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor[layer_id].layer_norm.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(__UpperCamelCase )
def __snake_case ( __UpperCamelCase : Optional[Any] ):
"""simple docstring"""
A_ , A_ = emb.weight.shape
A_ = nn.Linear(__UpperCamelCase ,__UpperCamelCase ,bias=__UpperCamelCase )
A_ = emb.weight.data
return lin_layer
def __snake_case ( __UpperCamelCase : Tuple ):
"""simple docstring"""
with open(__UpperCamelCase ,"r" ,encoding="utf-8" ) as f:
A_ = f.readlines()
A_ = [line.split(" " )[0] for line in lines]
A_ = len(__UpperCamelCase )
A_ = {
"<s>": 0,
"<pad>": 1,
"</s>": 2,
"<unk>": 3,
}
vocab_dict.update(dict(zip(__UpperCamelCase ,range(4 ,num_words + 4 ) ) ) )
return vocab_dict
@torch.no_grad()
def __snake_case ( __UpperCamelCase : List[str] ,__UpperCamelCase : Any ,__UpperCamelCase : List[Any] ,__UpperCamelCase : Union[str, Any] ,__UpperCamelCase : List[str] ,__UpperCamelCase : Optional[Any] ,__UpperCamelCase : Dict ,):
"""simple docstring"""
A_ = WavaVecaConfig.from_pretrained(__UpperCamelCase )
A_ = SpeechaTextaConfig.from_pretrained(
__UpperCamelCase ,vocab_size=__UpperCamelCase ,decoder_layers=__UpperCamelCase ,do_stable_layer_norm=__UpperCamelCase )
A_ = WavaVecaFeatureExtractor(
feature_size=1 ,sampling_rate=1_6000 ,padding_value=0 ,do_normalize=__UpperCamelCase ,return_attention_mask=__UpperCamelCase ,)
A_ , A_ , A_ = fairseq.checkpoint_utils.load_model_ensemble_and_task(
[checkpoint_path] ,arg_overrides={"data": "/".join(dict_path.split("/" )[:-1] )} )
A_ = model[0].eval()
# set weights for wav2vec2 encoder
A_ = WavaVecaModel(__UpperCamelCase )
A_ = recursively_load_weights_wavaveca(model.encoder ,__UpperCamelCase )
A_ = SpeechaTextaForCausalLM(__UpperCamelCase )
A_ , A_ = hf_decoder.model.decoder.load_state_dict(model.decoder.state_dict() ,strict=__UpperCamelCase )
# set output linear layer
unexpected_keys.remove("embed_out" )
A_ = nn.Parameter(model.decoder.embed_out.detach() )
# layer norm is init to identity matrix so leaving it is fine
logger.warning(f'''The following keys are missing when loading the decoder weights: {missing_keys}''' )
logger.warning(f'''The following keys are unexpected when loading the decoder weights: {unexpected_keys}''' )
A_ = SpeechEncoderDecoderModel(encoder=__UpperCamelCase ,decoder=__UpperCamelCase )
A_ = False
# add projection layer
A_ = nn.Parameter(projection_layer.weight )
A_ = nn.Parameter(projection_layer.bias )
A_ = create_vocab_dict(__UpperCamelCase )
with open(os.path.join(__UpperCamelCase ,"vocab.json" ) ,"w" ) as fp:
json.dump(__UpperCamelCase ,__UpperCamelCase )
A_ = SpeechaTextaTokenizer(os.path.join(__UpperCamelCase ,"vocab.json" ) )
tokenizer.save_pretrained(__UpperCamelCase )
A_ = hf_wavavec.config.to_dict()
A_ = tokenizer.pad_token_id
A_ = tokenizer.bos_token_id
A_ = tokenizer.eos_token_id
A_ = "speech_to_text_2"
A_ = "wav2vec2"
A_ = SpeechEncoderDecoderConfig.from_dict(__UpperCamelCase )
hf_wavavec.save_pretrained(__UpperCamelCase )
feature_extractor.save_pretrained(__UpperCamelCase )
if __name__ == "__main__":
__a :int = argparse.ArgumentParser()
parser.add_argument('--pytorch_dump_folder_path', default=None, type=str, help='Path to the output PyTorch model.')
parser.add_argument('--checkpoint_path', default=None, type=str, help='Path to fairseq checkpoint')
parser.add_argument('--dict_path', default=None, type=str, help='Path to dict of fine-tuned model')
parser.add_argument(
'--encoder_config_path',
default='facebook/wav2vec2-large-lv60',
type=str,
help='Path to hf encoder wav2vec2 checkpoint config',
)
parser.add_argument(
'--decoder_config_path',
default='facebook/s2t-small-mustc-en-fr-st',
type=str,
help='Path to hf decoder s2t checkpoint config',
)
parser.add_argument('--vocab_size', default=1_0224, type=int, help='Vocab size of decoder')
parser.add_argument('--num_decoder_layers', default=7, type=int, help='Number of decoder layers')
__a :Tuple = parser.parse_args()
convert_wavaveca_checkpoint(
args.checkpoint_path,
args.pytorch_dump_folder_path,
args.dict_path,
encoder_config_path=args.encoder_config_path,
decoder_config_path=args.decoder_config_path,
vocab_size=args.vocab_size,
num_decoder_layers=args.num_decoder_layers,
) | 86 | 0 |
'''simple docstring'''
import gc
import unittest
import numpy as np
import torch
from diffusers import StableDiffusionKDiffusionPipeline
from diffusers.utils import slow, torch_device
from diffusers.utils.testing_utils import enable_full_determinism, require_torch_gpu
enable_full_determinism()
@slow
@require_torch_gpu
class UpperCAmelCase_ (unittest.TestCase ):
"""simple docstring"""
def lowercase_ ( self ) -> Tuple:
# clean up the VRAM after each test
super().tearDown()
gc.collect()
torch.cuda.empty_cache()
def lowercase_ ( self ) -> Optional[Any]:
__lowerCamelCase : str = StableDiffusionKDiffusionPipeline.from_pretrained('CompVis/stable-diffusion-v1-4' )
__lowerCamelCase : Dict = sd_pipe.to(SCREAMING_SNAKE_CASE_ )
sd_pipe.set_progress_bar_config(disable=SCREAMING_SNAKE_CASE_ )
sd_pipe.set_scheduler('sample_euler' )
__lowerCamelCase : int = 'A painting of a squirrel eating a burger'
__lowerCamelCase : Tuple = torch.manual_seed(0 )
__lowerCamelCase : Tuple = sd_pipe([prompt] , generator=SCREAMING_SNAKE_CASE_ , guidance_scale=9.0 , num_inference_steps=20 , output_type='np' )
__lowerCamelCase : Union[str, Any] = output.images
__lowerCamelCase : List[Any] = image[0, -3:, -3:, -1]
assert image.shape == (1, 5_12, 5_12, 3)
__lowerCamelCase : List[str] = np.array([0.0_4_4_7, 0.0_4_9_2, 0.0_4_6_8, 0.0_4_0_8, 0.0_3_8_3, 0.0_4_0_8, 0.0_3_5_4, 0.0_3_8_0, 0.0_3_3_9] )
assert np.abs(image_slice.flatten() - expected_slice ).max() < 1E-2
def lowercase_ ( self ) -> Dict:
__lowerCamelCase : List[Any] = StableDiffusionKDiffusionPipeline.from_pretrained('stabilityai/stable-diffusion-2-1-base' )
__lowerCamelCase : List[Any] = sd_pipe.to(SCREAMING_SNAKE_CASE_ )
sd_pipe.set_progress_bar_config(disable=SCREAMING_SNAKE_CASE_ )
sd_pipe.set_scheduler('sample_euler' )
__lowerCamelCase : Union[str, Any] = 'A painting of a squirrel eating a burger'
__lowerCamelCase : Any = torch.manual_seed(0 )
__lowerCamelCase : List[str] = sd_pipe([prompt] , generator=SCREAMING_SNAKE_CASE_ , guidance_scale=9.0 , num_inference_steps=20 , output_type='np' )
__lowerCamelCase : List[str] = output.images
__lowerCamelCase : List[str] = image[0, -3:, -3:, -1]
assert image.shape == (1, 5_12, 5_12, 3)
__lowerCamelCase : Tuple = np.array([0.1_2_3_7, 0.1_3_2_0, 0.1_4_3_8, 0.1_3_5_9, 0.1_3_9_0, 0.1_1_3_2, 0.1_2_7_7, 0.1_1_7_5, 0.1_1_1_2] )
assert np.abs(image_slice.flatten() - expected_slice ).max() < 5E-1
def lowercase_ ( self ) -> List[Any]:
__lowerCamelCase : List[Any] = StableDiffusionKDiffusionPipeline.from_pretrained('stabilityai/stable-diffusion-2-1-base' )
__lowerCamelCase : Dict = sd_pipe.to(SCREAMING_SNAKE_CASE_ )
sd_pipe.set_progress_bar_config(disable=SCREAMING_SNAKE_CASE_ )
sd_pipe.set_scheduler('sample_dpmpp_2m' )
__lowerCamelCase : Any = 'A painting of a squirrel eating a burger'
__lowerCamelCase : Optional[Any] = torch.manual_seed(0 )
__lowerCamelCase : Optional[int] = sd_pipe(
[prompt] , generator=SCREAMING_SNAKE_CASE_ , guidance_scale=7.5 , num_inference_steps=15 , output_type='np' , use_karras_sigmas=SCREAMING_SNAKE_CASE_ , )
__lowerCamelCase : Optional[Any] = output.images
__lowerCamelCase : Any = image[0, -3:, -3:, -1]
assert image.shape == (1, 5_12, 5_12, 3)
__lowerCamelCase : List[Any] = np.array(
[0.1_1_3_8_1_6_8_9, 0.1_2_1_1_2_9_2_1, 0.1_3_8_9_4_5_7, 0.1_2_5_4_9_6_0_6, 0.1_2_4_4_9_6_4, 0.1_0_8_3_1_5_1_7, 0.1_1_5_6_2_8_6_6, 0.1_0_8_6_7_8_1_6, 0.1_0_4_9_9_0_4_8] )
assert np.abs(image_slice.flatten() - expected_slice ).max() < 1E-2
| 13 |
from typing import Any, Dict, List, Union
from ..utils import add_end_docstrings, is_torch_available, is_vision_available, logging, requires_backends
from .base import PIPELINE_INIT_ARGS, Pipeline
if is_vision_available():
from ..image_utils import load_image
if is_torch_available():
import torch
from ..models.auto.modeling_auto import MODEL_FOR_OBJECT_DETECTION_MAPPING, MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING
__a :str = logging.get_logger(__name__)
__a :Any = Dict[str, Any]
__a :int = List[Prediction]
@add_end_docstrings(snake_case_ )
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Tuple , *UpperCAmelCase : Optional[int] , **UpperCAmelCase : Optional[Any] ):
super().__init__(*UpperCAmelCase , **UpperCAmelCase )
if self.framework == "tf":
raise ValueError(f'''The {self.__class__} is only available in PyTorch.''' )
requires_backends(self , "vision" )
self.check_model_type(
dict(MODEL_FOR_OBJECT_DETECTION_MAPPING.items() + MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING.items() ) )
def __A ( self : str , **UpperCAmelCase : str ):
A_ = {}
if "threshold" in kwargs:
A_ = kwargs["threshold"]
return {}, {}, postprocess_kwargs
def __call__( self : Union[str, Any] , *UpperCAmelCase : Union[str, Any] , **UpperCAmelCase : Optional[Any] ):
return super().__call__(*UpperCAmelCase , **UpperCAmelCase )
def __A ( self : str , UpperCAmelCase : Any ):
A_ = load_image(UpperCAmelCase )
A_ = torch.IntTensor([[image.height, image.width]] )
A_ = self.image_processor(images=[image] , return_tensors="pt" )
if self.tokenizer is not None:
A_ = self.tokenizer(text=inputs["words"] , boxes=inputs["boxes"] , return_tensors="pt" )
A_ = target_size
return inputs
def __A ( self : Optional[Any] , UpperCAmelCase : Optional[int] ):
A_ = model_inputs.pop("target_size" )
A_ = self.model(**UpperCAmelCase )
A_ = outputs.__class__({"target_size": target_size, **outputs} )
if self.tokenizer is not None:
A_ = model_inputs["bbox"]
return model_outputs
def __A ( self : str , UpperCAmelCase : Optional[int] , UpperCAmelCase : Union[str, Any]=0.9 ):
A_ = model_outputs["target_size"]
if self.tokenizer is not None:
# This is a LayoutLMForTokenClassification variant.
# The OCR got the boxes and the model classified the words.
A_ , A_ = target_size[0].tolist()
def unnormalize(UpperCAmelCase : Any ):
return self._get_bounding_box(
torch.Tensor(
[
(width * bbox[0] / 1000),
(height * bbox[1] / 1000),
(width * bbox[2] / 1000),
(height * bbox[3] / 1000),
] ) )
A_ , A_ = model_outputs["logits"].squeeze(0 ).softmax(dim=-1 ).max(dim=-1 )
A_ = [self.model.config.idalabel[prediction] for prediction in classes.tolist()]
A_ = [unnormalize(UpperCAmelCase ) for bbox in model_outputs["bbox"].squeeze(0 )]
A_ = ["score", "label", "box"]
A_ = [dict(zip(UpperCAmelCase , UpperCAmelCase ) ) for vals in zip(scores.tolist() , UpperCAmelCase , UpperCAmelCase ) if vals[0] > threshold]
else:
# This is a regular ForObjectDetectionModel
A_ = self.image_processor.post_process_object_detection(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase )
A_ = raw_annotations[0]
A_ = raw_annotation["scores"]
A_ = raw_annotation["labels"]
A_ = raw_annotation["boxes"]
A_ = scores.tolist()
A_ = [self.model.config.idalabel[label.item()] for label in labels]
A_ = [self._get_bounding_box(UpperCAmelCase ) for box in boxes]
# {"scores": [...], ...} --> [{"score":x, ...}, ...]
A_ = ["score", "label", "box"]
A_ = [
dict(zip(UpperCAmelCase , UpperCAmelCase ) )
for vals in zip(raw_annotation["scores"] , raw_annotation["labels"] , raw_annotation["boxes"] )
]
return annotation
def __A ( self : Tuple , UpperCAmelCase : "torch.Tensor" ):
if self.framework != "pt":
raise ValueError("The ObjectDetectionPipeline is only available in PyTorch." )
A_ , A_ , A_ , A_ = box.int().tolist()
A_ = {
"xmin": xmin,
"ymin": ymin,
"xmax": xmax,
"ymax": ymax,
}
return bbox | 86 | 0 |
import gc
import random
import unittest
import numpy as np
import torch
from PIL import Image
from transformers import CLIPTextConfig, CLIPTextModel, CLIPTokenizer
from diffusers import AutoencoderKL, PNDMScheduler, StableDiffusionInpaintPipeline, UNetaDConditionModel
from diffusers.utils import floats_tensor, load_image, load_numpy, torch_device
from diffusers.utils.testing_utils import enable_full_determinism, require_torch_gpu, slow
from ..pipeline_params import TEXT_GUIDED_IMAGE_INPAINTING_BATCH_PARAMS, TEXT_GUIDED_IMAGE_INPAINTING_PARAMS
from ..test_pipelines_common import PipelineKarrasSchedulerTesterMixin, PipelineLatentTesterMixin, PipelineTesterMixin
enable_full_determinism()
class UpperCAmelCase_ ( __lowercase , __lowercase , __lowercase , unittest.TestCase ):
"""simple docstring"""
UpperCAmelCase__ : List[str] = StableDiffusionInpaintPipeline
UpperCAmelCase__ : Optional[Any] = TEXT_GUIDED_IMAGE_INPAINTING_PARAMS
UpperCAmelCase__ : int = TEXT_GUIDED_IMAGE_INPAINTING_BATCH_PARAMS
UpperCAmelCase__ : Dict = frozenset(
[] ) # TO-DO: update image_params once pipeline is refactored with VaeImageProcessor.preprocess
UpperCAmelCase__ : Optional[int] = frozenset([] )
def __lowercase ( self ) -> Tuple:
torch.manual_seed(0 )
_a : Tuple = UNetaDConditionModel(
block_out_channels=(3_2, 6_4) , layers_per_block=2 , sample_size=3_2 , in_channels=9 , out_channels=4 , down_block_types=('''DownBlock2D''', '''CrossAttnDownBlock2D''') , up_block_types=('''CrossAttnUpBlock2D''', '''UpBlock2D''') , cross_attention_dim=3_2 , attention_head_dim=(2, 4) , use_linear_projection=_a , )
_a : List[Any] = PNDMScheduler(skip_prk_steps=_a )
torch.manual_seed(0 )
_a : Tuple = AutoencoderKL(
block_out_channels=[3_2, 6_4] , in_channels=3 , out_channels=3 , down_block_types=['''DownEncoderBlock2D''', '''DownEncoderBlock2D'''] , up_block_types=['''UpDecoderBlock2D''', '''UpDecoderBlock2D'''] , latent_channels=4 , sample_size=1_2_8 , )
torch.manual_seed(0 )
_a : List[str] = CLIPTextConfig(
bos_token_id=0 , eos_token_id=2 , hidden_size=3_2 , intermediate_size=3_7 , layer_norm_eps=1e-0_5 , num_attention_heads=4 , num_hidden_layers=5 , pad_token_id=1 , vocab_size=1_0_0_0 , hidden_act='''gelu''' , projection_dim=5_1_2 , )
_a : str = CLIPTextModel(_a )
_a : str = CLIPTokenizer.from_pretrained('''hf-internal-testing/tiny-random-clip''' )
_a : Dict = {
'''unet''': unet,
'''scheduler''': scheduler,
'''vae''': vae,
'''text_encoder''': text_encoder,
'''tokenizer''': tokenizer,
'''safety_checker''': None,
'''feature_extractor''': None,
}
return components
def __lowercase ( self , _a , _a=0 ) -> List[Any]:
# TODO: use tensor inputs instead of PIL, this is here just to leave the old expected_slices untouched
_a : Union[str, Any] = floats_tensor((1, 3, 3_2, 3_2) , rng=random.Random(_a ) ).to(_a )
_a : str = image.cpu().permute(0 , 2 , 3 , 1 )[0]
_a : Tuple = Image.fromarray(np.uinta(_a ) ).convert('''RGB''' ).resize((6_4, 6_4) )
_a : Dict = Image.fromarray(np.uinta(image + 4 ) ).convert('''RGB''' ).resize((6_4, 6_4) )
if str(_a ).startswith('''mps''' ):
_a : Optional[int] = torch.manual_seed(_a )
else:
_a : Dict = torch.Generator(device=_a ).manual_seed(_a )
_a : Optional[int] = {
'''prompt''': '''A painting of a squirrel eating a burger''',
'''image''': init_image,
'''mask_image''': mask_image,
'''generator''': generator,
'''num_inference_steps''': 2,
'''guidance_scale''': 6.0,
'''output_type''': '''numpy''',
}
return inputs
def __lowercase ( self ) -> List[str]:
_a : int = '''cpu''' # ensure determinism for the device-dependent torch.Generator
_a : List[str] = self.get_dummy_components()
_a : List[str] = StableDiffusionInpaintPipeline(**_a )
_a : Tuple = sd_pipe.to(_a )
sd_pipe.set_progress_bar_config(disable=_a )
_a : Union[str, Any] = self.get_dummy_inputs(_a )
_a : List[str] = sd_pipe(**_a ).images
_a : Union[str, Any] = image[0, -3:, -3:, -1]
assert image.shape == (1, 6_4, 6_4, 3)
_a : str = np.array([0.4727, 0.5735, 0.3941, 0.5446, 0.5926, 0.4394, 0.5062, 0.4654, 0.4476] )
assert np.abs(image_slice.flatten() - expected_slice ).max() < 1e-2
def __lowercase ( self ) -> Any:
super().test_inference_batch_single_identical(expected_max_diff=3e-3 )
@slow
@require_torch_gpu
class UpperCAmelCase_ ( unittest.TestCase ):
"""simple docstring"""
def __lowercase ( self ) -> List[Any]:
# clean up the VRAM after each test
super().tearDown()
gc.collect()
torch.cuda.empty_cache()
def __lowercase ( self ) -> List[Any]:
_a : int = load_image(
'''https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main'''
'''/sd2-inpaint/init_image.png''' )
_a : Dict = load_image(
'''https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd2-inpaint/mask.png''' )
_a : Any = load_numpy(
'''https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd2-inpaint'''
'''/yellow_cat_sitting_on_a_park_bench.npy''' )
_a : Optional[int] = '''stabilityai/stable-diffusion-2-inpainting'''
_a : Union[str, Any] = StableDiffusionInpaintPipeline.from_pretrained(_a , safety_checker=_a )
pipe.to(_a )
pipe.set_progress_bar_config(disable=_a )
pipe.enable_attention_slicing()
_a : int = '''Face of a yellow cat, high resolution, sitting on a park bench'''
_a : List[Any] = torch.manual_seed(0 )
_a : Any = pipe(
prompt=_a , image=_a , mask_image=_a , generator=_a , output_type='''np''' , )
_a : Union[str, Any] = output.images[0]
assert image.shape == (5_1_2, 5_1_2, 3)
assert np.abs(expected_image - image ).max() < 9e-3
def __lowercase ( self ) -> Dict:
_a : List[Any] = load_image(
'''https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main'''
'''/sd2-inpaint/init_image.png''' )
_a : Optional[int] = load_image(
'''https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd2-inpaint/mask.png''' )
_a : List[Any] = load_numpy(
'''https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd2-inpaint'''
'''/yellow_cat_sitting_on_a_park_bench_fp16.npy''' )
_a : Tuple = '''stabilityai/stable-diffusion-2-inpainting'''
_a : Optional[int] = StableDiffusionInpaintPipeline.from_pretrained(
_a , torch_dtype=torch.floataa , safety_checker=_a , )
pipe.to(_a )
pipe.set_progress_bar_config(disable=_a )
pipe.enable_attention_slicing()
_a : List[Any] = '''Face of a yellow cat, high resolution, sitting on a park bench'''
_a : Dict = torch.manual_seed(0 )
_a : Any = pipe(
prompt=_a , image=_a , mask_image=_a , generator=_a , output_type='''np''' , )
_a : List[str] = output.images[0]
assert image.shape == (5_1_2, 5_1_2, 3)
assert np.abs(expected_image - image ).max() < 5e-1
def __lowercase ( self ) -> List[Any]:
torch.cuda.empty_cache()
torch.cuda.reset_max_memory_allocated()
torch.cuda.reset_peak_memory_stats()
_a : Any = load_image(
'''https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main'''
'''/sd2-inpaint/init_image.png''' )
_a : List[str] = load_image(
'''https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd2-inpaint/mask.png''' )
_a : List[Any] = '''stabilityai/stable-diffusion-2-inpainting'''
_a : Optional[int] = PNDMScheduler.from_pretrained(_a , subfolder='''scheduler''' )
_a : Tuple = StableDiffusionInpaintPipeline.from_pretrained(
_a , safety_checker=_a , scheduler=_a , torch_dtype=torch.floataa , )
pipe.to(_a )
pipe.set_progress_bar_config(disable=_a )
pipe.enable_attention_slicing(1 )
pipe.enable_sequential_cpu_offload()
_a : Tuple = '''Face of a yellow cat, high resolution, sitting on a park bench'''
_a : Tuple = torch.manual_seed(0 )
_a : Dict = pipe(
prompt=_a , image=_a , mask_image=_a , generator=_a , num_inference_steps=2 , output_type='''np''' , )
_a : List[str] = torch.cuda.max_memory_allocated()
# make sure that less than 2.65 GB is allocated
assert mem_bytes < 2.65 * 1_0**9
| 14 |
import inspect
from typing import List, Optional, Tuple, Union
import numpy as np
import PIL
import torch
import torch.utils.checkpoint
from ...models import UNetaDModel, VQModel
from ...schedulers import (
DDIMScheduler,
DPMSolverMultistepScheduler,
EulerAncestralDiscreteScheduler,
EulerDiscreteScheduler,
LMSDiscreteScheduler,
PNDMScheduler,
)
from ...utils import PIL_INTERPOLATION, randn_tensor
from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
def __snake_case ( __UpperCamelCase : Dict ):
"""simple docstring"""
A_ , A_ = image.size
A_ , A_ = (x - x % 32 for x in (w, h)) # resize to integer multiple of 32
A_ = image.resize((w, h) ,resample=PIL_INTERPOLATION["lanczos"] )
A_ = np.array(__UpperCamelCase ).astype(np.floataa ) / 255.0
A_ = image[None].transpose(0 ,3 ,1 ,2 )
A_ = torch.from_numpy(__UpperCamelCase )
return 2.0 * image - 1.0
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Union[str, Any] , UpperCAmelCase : VQModel , UpperCAmelCase : UNetaDModel , UpperCAmelCase : Union[
DDIMScheduler,
PNDMScheduler,
LMSDiscreteScheduler,
EulerDiscreteScheduler,
EulerAncestralDiscreteScheduler,
DPMSolverMultistepScheduler,
] , ):
super().__init__()
self.register_modules(vqvae=UpperCAmelCase , unet=UpperCAmelCase , scheduler=UpperCAmelCase )
@torch.no_grad()
def __call__( self : int , UpperCAmelCase : Union[torch.Tensor, PIL.Image.Image] = None , UpperCAmelCase : Optional[int] = 1 , UpperCAmelCase : Optional[int] = 100 , UpperCAmelCase : Optional[float] = 0.0 , UpperCAmelCase : Optional[Union[torch.Generator, List[torch.Generator]]] = None , UpperCAmelCase : Optional[str] = "pil" , UpperCAmelCase : bool = True , ):
if isinstance(UpperCAmelCase , PIL.Image.Image ):
A_ = 1
elif isinstance(UpperCAmelCase , torch.Tensor ):
A_ = image.shape[0]
else:
raise ValueError(f'''`image` has to be of type `PIL.Image.Image` or `torch.Tensor` but is {type(UpperCAmelCase )}''' )
if isinstance(UpperCAmelCase , PIL.Image.Image ):
A_ = preprocess(UpperCAmelCase )
A_ , A_ = image.shape[-2:]
# in_channels should be 6: 3 for latents, 3 for low resolution image
A_ = (batch_size, self.unet.config.in_channels // 2, height, width)
A_ = next(self.unet.parameters() ).dtype
A_ = randn_tensor(UpperCAmelCase , generator=UpperCAmelCase , device=self.device , dtype=UpperCAmelCase )
A_ = image.to(device=self.device , dtype=UpperCAmelCase )
# set timesteps and move to the correct device
self.scheduler.set_timesteps(UpperCAmelCase , device=self.device )
A_ = self.scheduler.timesteps
# scale the initial noise by the standard deviation required by the scheduler
A_ = latents * self.scheduler.init_noise_sigma
# prepare extra kwargs for the scheduler step, since not all schedulers have the same signature.
# eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
# eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
# and should be between [0, 1]
A_ = "eta" in set(inspect.signature(self.scheduler.step ).parameters.keys() )
A_ = {}
if accepts_eta:
A_ = eta
for t in self.progress_bar(UpperCAmelCase ):
# concat latents and low resolution image in the channel dimension.
A_ = torch.cat([latents, image] , dim=1 )
A_ = self.scheduler.scale_model_input(UpperCAmelCase , UpperCAmelCase )
# predict the noise residual
A_ = self.unet(UpperCAmelCase , UpperCAmelCase ).sample
# compute the previous noisy sample x_t -> x_t-1
A_ = self.scheduler.step(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase , **UpperCAmelCase ).prev_sample
# decode the image latents with the VQVAE
A_ = self.vqvae.decode(UpperCAmelCase ).sample
A_ = torch.clamp(UpperCAmelCase , -1.0 , 1.0 )
A_ = image / 2 + 0.5
A_ = image.cpu().permute(0 , 2 , 3 , 1 ).numpy()
if output_type == "pil":
A_ = self.numpy_to_pil(UpperCAmelCase )
if not return_dict:
return (image,)
return ImagePipelineOutput(images=UpperCAmelCase ) | 86 | 0 |
import argparse
import json
import os
import pickle
import shutil
import numpy as np
import torch
from distiller import Distiller
from lm_seqs_dataset import LmSeqsDataset
from transformers import (
BertConfig,
BertForMaskedLM,
BertTokenizer,
DistilBertConfig,
DistilBertForMaskedLM,
DistilBertTokenizer,
GPTaConfig,
GPTaLMHeadModel,
GPTaTokenizer,
RobertaConfig,
RobertaForMaskedLM,
RobertaTokenizer,
)
from utils import git_log, init_gpu_params, logger, set_seed
A : int = {
'distilbert': (DistilBertConfig, DistilBertForMaskedLM, DistilBertTokenizer),
'roberta': (RobertaConfig, RobertaForMaskedLM, RobertaTokenizer),
'bert': (BertConfig, BertForMaskedLM, BertTokenizer),
'gpt2': (GPTaConfig, GPTaLMHeadModel, GPTaTokenizer),
}
def UpperCamelCase ( __magic_name__ : List[str] ) -> Union[str, Any]:
"""simple docstring"""
assert (args.mlm and args.alpha_mlm > 0.0) or (not args.mlm and args.alpha_mlm == 0.0)
assert (args.alpha_mlm > 0.0 and args.alpha_clm == 0.0) or (args.alpha_mlm == 0.0 and args.alpha_clm > 0.0)
if args.mlm:
assert os.path.isfile(args.token_counts )
assert (args.student_type in ["roberta", "distilbert"]) and (args.teacher_type in ["roberta", "bert"])
else:
assert (args.student_type in ["gpt2"]) and (args.teacher_type in ["gpt2"])
assert args.teacher_type == args.student_type or (
args.student_type == "distilbert" and args.teacher_type == "bert"
)
assert os.path.isfile(args.student_config )
if args.student_pretrained_weights is not None:
assert os.path.isfile(args.student_pretrained_weights )
if args.freeze_token_type_embds:
assert args.student_type in ["roberta"]
assert args.alpha_ce >= 0.0
assert args.alpha_mlm >= 0.0
assert args.alpha_clm >= 0.0
assert args.alpha_mse >= 0.0
assert args.alpha_cos >= 0.0
assert args.alpha_ce + args.alpha_mlm + args.alpha_clm + args.alpha_mse + args.alpha_cos > 0.0
def UpperCamelCase ( __magic_name__ : int , __magic_name__ : str ) -> Optional[int]:
"""simple docstring"""
if args.student_type == "roberta":
lowercase__ = False
elif args.student_type == "gpt2":
lowercase__ = False
def UpperCamelCase ( __magic_name__ : Tuple , __magic_name__ : Tuple ) -> Tuple:
"""simple docstring"""
if args.student_type == "roberta":
lowercase__ = False
def UpperCamelCase ( ) -> str:
"""simple docstring"""
lowercase__ = argparse.ArgumentParser(description="""Training""" )
parser.add_argument("""--force""" , action="""store_true""" , help="""Overwrite dump_path if it already exists.""" )
parser.add_argument(
"""--dump_path""" , type=__magic_name__ , required=__magic_name__ , help="""The output directory (log, checkpoints, parameters, etc.)""" )
parser.add_argument(
"""--data_file""" , type=__magic_name__ , required=__magic_name__ , help="""The binarized file (tokenized + tokens_to_ids) and grouped by sequence.""" , )
parser.add_argument(
"""--student_type""" , type=__magic_name__ , choices=["""distilbert""", """roberta""", """gpt2"""] , required=__magic_name__ , help="""The student type (DistilBERT, RoBERTa).""" , )
parser.add_argument("""--student_config""" , type=__magic_name__ , required=__magic_name__ , help="""Path to the student configuration.""" )
parser.add_argument(
"""--student_pretrained_weights""" , default=__magic_name__ , type=__magic_name__ , help="""Load student initialization checkpoint.""" )
parser.add_argument(
"""--teacher_type""" , choices=["""bert""", """roberta""", """gpt2"""] , required=__magic_name__ , help="""Teacher type (BERT, RoBERTa).""" )
parser.add_argument("""--teacher_name""" , type=__magic_name__ , required=__magic_name__ , help="""The teacher model.""" )
parser.add_argument("""--temperature""" , default=2.0 , type=__magic_name__ , help="""Temperature for the softmax temperature.""" )
parser.add_argument(
"""--alpha_ce""" , default=0.5 , type=__magic_name__ , help="""Linear weight for the distillation loss. Must be >=0.""" )
parser.add_argument(
"""--alpha_mlm""" , default=0.0 , type=__magic_name__ , help="""Linear weight for the MLM loss. Must be >=0. Should be used in conjunction with `mlm` flag.""" , )
parser.add_argument("""--alpha_clm""" , default=0.5 , type=__magic_name__ , help="""Linear weight for the CLM loss. Must be >=0.""" )
parser.add_argument("""--alpha_mse""" , default=0.0 , type=__magic_name__ , help="""Linear weight of the MSE loss. Must be >=0.""" )
parser.add_argument(
"""--alpha_cos""" , default=0.0 , type=__magic_name__ , help="""Linear weight of the cosine embedding loss. Must be >=0.""" )
parser.add_argument(
"""--mlm""" , action="""store_true""" , help="""The LM step: MLM or CLM. If `mlm` is True, the MLM is used over CLM.""" )
parser.add_argument(
"""--mlm_mask_prop""" , default=0.1_5 , type=__magic_name__ , help="""Proportion of tokens for which we need to make a prediction.""" , )
parser.add_argument("""--word_mask""" , default=0.8 , type=__magic_name__ , help="""Proportion of tokens to mask out.""" )
parser.add_argument("""--word_keep""" , default=0.1 , type=__magic_name__ , help="""Proportion of tokens to keep.""" )
parser.add_argument("""--word_rand""" , default=0.1 , type=__magic_name__ , help="""Proportion of tokens to randomly replace.""" )
parser.add_argument(
"""--mlm_smoothing""" , default=0.7 , type=__magic_name__ , help="""Smoothing parameter to emphasize more rare tokens (see XLM, similar to word2vec).""" , )
parser.add_argument("""--token_counts""" , type=__magic_name__ , help="""The token counts in the data_file for MLM.""" )
parser.add_argument(
"""--restrict_ce_to_mask""" , action="""store_true""" , help="""If true, compute the distillation loss only the [MLM] prediction distribution.""" , )
parser.add_argument(
"""--freeze_pos_embs""" , action="""store_true""" , help="""Freeze positional embeddings during distillation. For student_type in ['roberta', 'gpt2'] only.""" , )
parser.add_argument(
"""--freeze_token_type_embds""" , action="""store_true""" , help="""Freeze token type embeddings during distillation if existent. For student_type in ['roberta'] only.""" , )
parser.add_argument("""--n_epoch""" , type=__magic_name__ , default=3 , help="""Number of pass on the whole dataset.""" )
parser.add_argument("""--batch_size""" , type=__magic_name__ , default=5 , help="""Batch size (for each process).""" )
parser.add_argument(
"""--group_by_size""" , action="""store_false""" , help="""If true, group sequences that have similar length into the same batch. Default is true.""" , )
parser.add_argument(
"""--gradient_accumulation_steps""" , type=__magic_name__ , default=50 , help="""Gradient accumulation for larger training batches.""" , )
parser.add_argument("""--warmup_prop""" , default=0.0_5 , type=__magic_name__ , help="""Linear warmup proportion.""" )
parser.add_argument("""--weight_decay""" , default=0.0 , type=__magic_name__ , help="""Weight decay if we apply some.""" )
parser.add_argument("""--learning_rate""" , default=5E-4 , type=__magic_name__ , help="""The initial learning rate for Adam.""" )
parser.add_argument("""--adam_epsilon""" , default=1E-6 , type=__magic_name__ , help="""Epsilon for Adam optimizer.""" )
parser.add_argument("""--max_grad_norm""" , default=5.0 , type=__magic_name__ , help="""Max gradient norm.""" )
parser.add_argument("""--initializer_range""" , default=0.0_2 , type=__magic_name__ , help="""Random initialization range.""" )
parser.add_argument(
"""--fp16""" , action="""store_true""" , help="""Whether to use 16-bit (mixed) precision (through NVIDIA apex) instead of 32-bit""" , )
parser.add_argument(
"""--fp16_opt_level""" , type=__magic_name__ , default="""O1""" , help=(
"""For fp16: Apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3']."""
"""See details at https://nvidia.github.io/apex/amp.html"""
) , )
parser.add_argument("""--n_gpu""" , type=__magic_name__ , default=1 , help="""Number of GPUs in the node.""" )
parser.add_argument("""--local_rank""" , type=__magic_name__ , default=-1 , help="""Distributed training - Local rank""" )
parser.add_argument("""--seed""" , type=__magic_name__ , default=56 , help="""Random seed""" )
parser.add_argument("""--log_interval""" , type=__magic_name__ , default=500 , help="""Tensorboard logging interval.""" )
parser.add_argument("""--checkpoint_interval""" , type=__magic_name__ , default=4000 , help="""Checkpoint interval.""" )
lowercase__ = parser.parse_args()
sanity_checks(__magic_name__ )
# ARGS #
init_gpu_params(__magic_name__ )
set_seed(__magic_name__ )
if args.is_master:
if os.path.exists(args.dump_path ):
if not args.force:
raise ValueError(
f'''Serialization dir {args.dump_path} already exists, but you have not precised wheter to overwrite'''
""" itUse `--force` if you want to overwrite it""" )
else:
shutil.rmtree(args.dump_path )
if not os.path.exists(args.dump_path ):
os.makedirs(args.dump_path )
logger.info(f'''Experiment will be dumped and logged in {args.dump_path}''' )
# SAVE PARAMS #
logger.info(f'''Param: {args}''' )
with open(os.path.join(args.dump_path , """parameters.json""" ) , """w""" ) as f:
json.dump(vars(__magic_name__ ) , __magic_name__ , indent=4 )
git_log(args.dump_path )
lowercase__ , lowercase__ , lowercase__ = MODEL_CLASSES[args.student_type]
lowercase__ , lowercase__ , lowercase__ = MODEL_CLASSES[args.teacher_type]
# TOKENIZER #
lowercase__ = teacher_tokenizer_class.from_pretrained(args.teacher_name )
lowercase__ = {}
for tok_name, tok_symbol in tokenizer.special_tokens_map.items():
lowercase__ = tokenizer.all_special_tokens.index(__magic_name__ )
lowercase__ = tokenizer.all_special_ids[idx]
logger.info(f'''Special tokens {special_tok_ids}''' )
lowercase__ = special_tok_ids
lowercase__ = tokenizer.max_model_input_sizes[args.teacher_name]
# DATA LOADER #
logger.info(f'''Loading data from {args.data_file}''' )
with open(args.data_file , """rb""" ) as fp:
lowercase__ = pickle.load(__magic_name__ )
if args.mlm:
logger.info(f'''Loading token counts from {args.token_counts} (already pre-computed)''' )
with open(args.token_counts , """rb""" ) as fp:
lowercase__ = pickle.load(__magic_name__ )
lowercase__ = np.maximum(__magic_name__ , 1 ) ** -args.mlm_smoothing
for idx in special_tok_ids.values():
lowercase__ = 0.0 # do not predict special tokens
lowercase__ = torch.from_numpy(__magic_name__ )
else:
lowercase__ = None
lowercase__ = LmSeqsDataset(params=__magic_name__ , data=__magic_name__ )
logger.info("""Data loader created.""" )
# STUDENT #
logger.info(f'''Loading student config from {args.student_config}''' )
lowercase__ = student_config_class.from_pretrained(args.student_config )
lowercase__ = True
if args.student_pretrained_weights is not None:
logger.info(f'''Loading pretrained weights from {args.student_pretrained_weights}''' )
lowercase__ = student_model_class.from_pretrained(args.student_pretrained_weights , config=__magic_name__ )
else:
lowercase__ = student_model_class(__magic_name__ )
if args.n_gpu > 0:
student.to(f'''cuda:{args.local_rank}''' )
logger.info("""Student loaded.""" )
# TEACHER #
lowercase__ = teacher_model_class.from_pretrained(args.teacher_name , output_hidden_states=__magic_name__ )
if args.n_gpu > 0:
teacher.to(f'''cuda:{args.local_rank}''' )
logger.info(f'''Teacher loaded from {args.teacher_name}.''' )
# FREEZING #
if args.freeze_pos_embs:
freeze_pos_embeddings(__magic_name__ , __magic_name__ )
if args.freeze_token_type_embds:
freeze_token_type_embeddings(__magic_name__ , __magic_name__ )
# SANITY CHECKS #
assert student.config.vocab_size == teacher.config.vocab_size
assert student.config.hidden_size == teacher.config.hidden_size
assert student.config.max_position_embeddings == teacher.config.max_position_embeddings
if args.mlm:
assert token_probs.size(0 ) == stu_architecture_config.vocab_size
# DISTILLER #
torch.cuda.empty_cache()
lowercase__ = Distiller(
params=__magic_name__ , dataset=__magic_name__ , token_probs=__magic_name__ , student=__magic_name__ , teacher=__magic_name__ )
distiller.train()
logger.info("""Let's go get some drinks.""" )
if __name__ == "__main__":
main()
| 15 |
__a :Optional[int] = [sum(int(c, 10) ** 2 for c in i.__str__()) for i in range(10_0000)]
def __snake_case ( __UpperCamelCase : int ):
"""simple docstring"""
A_ = 0
while number:
# Increased Speed Slightly by checking every 5 digits together.
sum_of_digits_squared += DIGITS_SQUARED[number % 10_0000]
number //= 10_0000
return sum_of_digits_squared
# There are 2 Chains made,
# One ends with 89 with the chain member 58 being the one which when declared first,
# there will be the least number of iterations for all the members to be checked.
# The other one ends with 1 and has only one element 1.
# So 58 and 1 are chosen to be declared at the starting.
# Changed dictionary to an array to quicken the solution
__a :list[bool | None] = [None] * 1000_0000
__a :Optional[Any] = True
__a :List[Any] = False
def __snake_case ( __UpperCamelCase : int ):
"""simple docstring"""
if CHAINS[number - 1] is not None:
return CHAINS[number - 1] # type: ignore
A_ = chain(next_number(__UpperCamelCase ) )
A_ = number_chain
while number < 1000_0000:
A_ = number_chain
number *= 10
return number_chain
def __snake_case ( __UpperCamelCase : int = 1000_0000 ):
"""simple docstring"""
for i in range(1 ,__UpperCamelCase ):
if CHAINS[i] is None:
chain(i + 1 )
return CHAINS[:number].count(__UpperCamelCase )
if __name__ == "__main__":
import doctest
doctest.testmod()
print(F"{solution() = }") | 86 | 0 |
import copy
import inspect
import unittest
import numpy as np
from huggingface_hub import hf_hub_download
from transformers import TimesformerConfig
from transformers.models.auto import get_values
from transformers.testing_utils import require_torch, require_vision, slow, torch_device
from transformers.utils import cached_property, is_torch_available, is_vision_available
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from torch import nn
from transformers import (
MODEL_FOR_VIDEO_CLASSIFICATION_MAPPING,
TimesformerForVideoClassification,
TimesformerModel,
)
from transformers.models.timesformer.modeling_timesformer import TIMESFORMER_PRETRAINED_MODEL_ARCHIVE_LIST
if is_vision_available():
from transformers import VideoMAEImageProcessor
class _SCREAMING_SNAKE_CASE :
'''simple docstring'''
def __init__( self : List[str] , __lowerCamelCase : Union[str, Any] , __lowerCamelCase : str=13 , __lowerCamelCase : List[Any]=10 , __lowerCamelCase : Optional[Any]=3 , __lowerCamelCase : int=2 , __lowerCamelCase : List[str]=2 , __lowerCamelCase : List[str]=True , __lowerCamelCase : Optional[int]=True , __lowerCamelCase : List[str]=32 , __lowerCamelCase : Dict=5 , __lowerCamelCase : Union[str, Any]=4 , __lowerCamelCase : List[Any]=37 , __lowerCamelCase : Tuple="gelu" , __lowerCamelCase : Any=0.1 , __lowerCamelCase : Tuple=0.1 , __lowerCamelCase : Any=10 , __lowerCamelCase : List[Any]=0.02 , __lowerCamelCase : List[str]="divided_space_time" , __lowerCamelCase : List[Any]=None , ):
SCREAMING_SNAKE_CASE = parent
SCREAMING_SNAKE_CASE = batch_size
SCREAMING_SNAKE_CASE = image_size
SCREAMING_SNAKE_CASE = num_channels
SCREAMING_SNAKE_CASE = patch_size
SCREAMING_SNAKE_CASE = num_frames
SCREAMING_SNAKE_CASE = is_training
SCREAMING_SNAKE_CASE = use_labels
SCREAMING_SNAKE_CASE = hidden_size
SCREAMING_SNAKE_CASE = num_hidden_layers
SCREAMING_SNAKE_CASE = num_attention_heads
SCREAMING_SNAKE_CASE = intermediate_size
SCREAMING_SNAKE_CASE = hidden_act
SCREAMING_SNAKE_CASE = hidden_dropout_prob
SCREAMING_SNAKE_CASE = attention_probs_dropout_prob
SCREAMING_SNAKE_CASE = attention_type
SCREAMING_SNAKE_CASE = initializer_range
SCREAMING_SNAKE_CASE = scope
SCREAMING_SNAKE_CASE = num_labels
# in TimeSformer, the number of spatial tokens equals num_frames * num_patches per frame + 1 CLS token
SCREAMING_SNAKE_CASE = (image_size // patch_size) ** 2
SCREAMING_SNAKE_CASE = (num_frames) * self.num_patches_per_frame + 1
def _snake_case ( self : int ):
SCREAMING_SNAKE_CASE = floats_tensor(
[self.batch_size, self.num_frames, self.num_channels, self.image_size, self.image_size] )
SCREAMING_SNAKE_CASE = None
if self.use_labels:
SCREAMING_SNAKE_CASE = ids_tensor([self.batch_size] , self.num_labels )
SCREAMING_SNAKE_CASE = self.get_config()
return config, pixel_values, labels
def _snake_case ( self : Dict ):
SCREAMING_SNAKE_CASE = TimesformerConfig(
image_size=self.image_size , patch_size=self.patch_size , num_channels=self.num_channels , num_frames=self.num_frames , hidden_size=self.hidden_size , num_hidden_layers=self.num_hidden_layers , num_attention_heads=self.num_attention_heads , intermediate_size=self.intermediate_size , hidden_act=self.hidden_act , hidden_dropout_prob=self.hidden_dropout_prob , attention_probs_dropout_prob=self.attention_probs_dropout_prob , initializer_range=self.initializer_range , attention_type=self.attention_type , )
SCREAMING_SNAKE_CASE = self.num_labels
return config
def _snake_case ( self : str , __lowerCamelCase : Optional[int] , __lowerCamelCase : Tuple , __lowerCamelCase : Optional[Any] ):
SCREAMING_SNAKE_CASE = TimesformerModel(config=__lowerCamelCase )
model.to(__lowerCamelCase )
model.eval()
SCREAMING_SNAKE_CASE = model(__lowerCamelCase )
self.parent.assertEqual(result.last_hidden_state.shape , (self.batch_size, self.seq_length, self.hidden_size) )
def _snake_case ( self : Optional[int] , __lowerCamelCase : Union[str, Any] , __lowerCamelCase : List[Any] , __lowerCamelCase : Optional[Any] ):
SCREAMING_SNAKE_CASE = TimesformerForVideoClassification(__lowerCamelCase )
model.to(__lowerCamelCase )
model.eval()
SCREAMING_SNAKE_CASE = model(__lowerCamelCase )
# verify the logits shape
SCREAMING_SNAKE_CASE = torch.Size((self.batch_size, self.num_labels) )
self.parent.assertEqual(result.logits.shape , __lowerCamelCase )
def _snake_case ( self : Tuple ):
SCREAMING_SNAKE_CASE = self.prepare_config_and_inputs()
SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE = config_and_inputs
SCREAMING_SNAKE_CASE = {"pixel_values": pixel_values}
return config, inputs_dict
@require_torch
class _SCREAMING_SNAKE_CASE ( __snake_case , __snake_case , unittest.TestCase ):
'''simple docstring'''
lowerCamelCase__ = (TimesformerModel, TimesformerForVideoClassification) if is_torch_available() else ()
lowerCamelCase__ = (
{"feature-extraction": TimesformerModel, "video-classification": TimesformerForVideoClassification}
if is_torch_available()
else {}
)
lowerCamelCase__ = False
lowerCamelCase__ = False
lowerCamelCase__ = False
lowerCamelCase__ = False
def _snake_case ( self : Optional[Any] ):
SCREAMING_SNAKE_CASE = TimesformerModelTester(self )
SCREAMING_SNAKE_CASE = ConfigTester(
self , config_class=__lowerCamelCase , has_text_modality=__lowerCamelCase , hidden_size=37 )
def _snake_case ( self : Union[str, Any] , __lowerCamelCase : Any , __lowerCamelCase : str , __lowerCamelCase : str=False ):
SCREAMING_SNAKE_CASE = copy.deepcopy(__lowerCamelCase )
if return_labels:
if model_class in get_values(__lowerCamelCase ):
SCREAMING_SNAKE_CASE = torch.zeros(
self.model_tester.batch_size , dtype=torch.long , device=__lowerCamelCase )
return inputs_dict
def _snake_case ( self : int ):
self.config_tester.run_common_tests()
@unittest.skip(reason="TimeSformer does not use inputs_embeds" )
def _snake_case ( self : Optional[Any] ):
pass
def _snake_case ( self : int ):
SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
SCREAMING_SNAKE_CASE = model_class(__lowerCamelCase )
self.assertIsInstance(model.get_input_embeddings() , (nn.Module) )
SCREAMING_SNAKE_CASE = model.get_output_embeddings()
self.assertTrue(x is None or isinstance(__lowerCamelCase , nn.Linear ) )
def _snake_case ( self : Optional[int] ):
SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
SCREAMING_SNAKE_CASE = model_class(__lowerCamelCase )
SCREAMING_SNAKE_CASE = inspect.signature(model.forward )
# signature.parameters is an OrderedDict => so arg_names order is deterministic
SCREAMING_SNAKE_CASE = [*signature.parameters.keys()]
SCREAMING_SNAKE_CASE = ["pixel_values"]
self.assertListEqual(arg_names[:1] , __lowerCamelCase )
def _snake_case ( self : Any ):
SCREAMING_SNAKE_CASE = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*__lowerCamelCase )
def _snake_case ( self : Dict ):
SCREAMING_SNAKE_CASE = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_video_classification(*__lowerCamelCase )
@slow
def _snake_case ( self : str ):
for model_name in TIMESFORMER_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
SCREAMING_SNAKE_CASE = TimesformerModel.from_pretrained(__lowerCamelCase )
self.assertIsNotNone(__lowerCamelCase )
def _snake_case ( self : List[Any] ):
if not self.has_attentions:
pass
else:
SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE = self.model_tester.prepare_config_and_inputs_for_common()
SCREAMING_SNAKE_CASE = True
for model_class in self.all_model_classes:
SCREAMING_SNAKE_CASE = self.model_tester.seq_length
SCREAMING_SNAKE_CASE = self.model_tester.num_frames
SCREAMING_SNAKE_CASE = True
SCREAMING_SNAKE_CASE = False
SCREAMING_SNAKE_CASE = True
SCREAMING_SNAKE_CASE = model_class(__lowerCamelCase )
model.to(__lowerCamelCase )
model.eval()
with torch.no_grad():
SCREAMING_SNAKE_CASE = model(**self._prepare_for_class(__lowerCamelCase , __lowerCamelCase ) )
SCREAMING_SNAKE_CASE = outputs.attentions
self.assertEqual(len(__lowerCamelCase ) , self.model_tester.num_hidden_layers )
# check that output_attentions also work using config
del inputs_dict["output_attentions"]
SCREAMING_SNAKE_CASE = True
SCREAMING_SNAKE_CASE = model_class(__lowerCamelCase )
model.to(__lowerCamelCase )
model.eval()
with torch.no_grad():
SCREAMING_SNAKE_CASE = model(**self._prepare_for_class(__lowerCamelCase , __lowerCamelCase ) )
SCREAMING_SNAKE_CASE = outputs.attentions
self.assertEqual(len(__lowerCamelCase ) , self.model_tester.num_hidden_layers )
# attentions has shape (batch_size x num_frames) x num_heads x (num_patches per frame + 1) x (num_patches per frame + 1)
self.assertListEqual(
list(attentions[0].shape[-3:] ) , [self.model_tester.num_attention_heads, seq_len // num_frames + 1, seq_len // num_frames + 1] , )
SCREAMING_SNAKE_CASE = len(__lowerCamelCase )
# Check attention is always last and order is fine
SCREAMING_SNAKE_CASE = True
SCREAMING_SNAKE_CASE = True
SCREAMING_SNAKE_CASE = model_class(__lowerCamelCase )
model.to(__lowerCamelCase )
model.eval()
with torch.no_grad():
SCREAMING_SNAKE_CASE = model(**self._prepare_for_class(__lowerCamelCase , __lowerCamelCase ) )
self.assertEqual(out_len + 1 , len(__lowerCamelCase ) )
SCREAMING_SNAKE_CASE = outputs.attentions
self.assertEqual(len(__lowerCamelCase ) , self.model_tester.num_hidden_layers )
# attentions has shape (batch_size x num_frames) x num_heads x (num_patches per frame + 1) x (num_patches per frame + 1)
self.assertListEqual(
list(self_attentions[0].shape[-3:] ) , [self.model_tester.num_attention_heads, seq_len // num_frames + 1, seq_len // num_frames + 1] , )
def _snake_case ( self : Any ):
def check_hidden_states_output(__lowerCamelCase : Optional[Any] , __lowerCamelCase : Optional[Any] , __lowerCamelCase : Dict ):
SCREAMING_SNAKE_CASE = model_class(__lowerCamelCase )
model.to(__lowerCamelCase )
model.eval()
with torch.no_grad():
SCREAMING_SNAKE_CASE = model(**self._prepare_for_class(__lowerCamelCase , __lowerCamelCase ) )
SCREAMING_SNAKE_CASE = outputs.hidden_states
SCREAMING_SNAKE_CASE = self.model_tester.num_hidden_layers + 1
self.assertEqual(len(__lowerCamelCase ) , __lowerCamelCase )
SCREAMING_SNAKE_CASE = self.model_tester.seq_length
self.assertListEqual(
list(hidden_states[0].shape[-2:] ) , [seq_length, self.model_tester.hidden_size] , )
SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
SCREAMING_SNAKE_CASE = True
check_hidden_states_output(__lowerCamelCase , __lowerCamelCase , __lowerCamelCase )
# check that output_hidden_states also work using config
del inputs_dict["output_hidden_states"]
SCREAMING_SNAKE_CASE = True
check_hidden_states_output(__lowerCamelCase , __lowerCamelCase , __lowerCamelCase )
def __a ( ):
SCREAMING_SNAKE_CASE = hf_hub_download(
repo_id="hf-internal-testing/spaghetti-video" , filename="eating_spaghetti.npy" , repo_type="dataset" )
SCREAMING_SNAKE_CASE = np.load(A__ )
return list(A__ )
@require_torch
@require_vision
class _SCREAMING_SNAKE_CASE ( unittest.TestCase ):
'''simple docstring'''
@cached_property
def _snake_case ( self : Union[str, Any] ):
# logits were tested with a different mean and std, so we use the same here
return (
VideoMAEImageProcessor(image_mean=[0.5, 0.5, 0.5] , image_std=[0.5, 0.5, 0.5] )
if is_vision_available()
else None
)
@slow
def _snake_case ( self : Any ):
SCREAMING_SNAKE_CASE = TimesformerForVideoClassification.from_pretrained("facebook/timesformer-base-finetuned-k400" ).to(
__lowerCamelCase )
SCREAMING_SNAKE_CASE = self.default_image_processor
SCREAMING_SNAKE_CASE = prepare_video()
SCREAMING_SNAKE_CASE = image_processor(video[:8] , return_tensors="pt" ).to(__lowerCamelCase )
# forward pass
with torch.no_grad():
SCREAMING_SNAKE_CASE = model(**__lowerCamelCase )
# verify the logits
SCREAMING_SNAKE_CASE = torch.Size((1, 400) )
self.assertEqual(outputs.logits.shape , __lowerCamelCase )
SCREAMING_SNAKE_CASE = torch.tensor([-0.3_016, -0.7_713, -0.4_205] ).to(__lowerCamelCase )
self.assertTrue(torch.allclose(outputs.logits[0, :3] , __lowerCamelCase , atol=1e-4 ) ) | 16 |
from typing import TYPE_CHECKING
from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_tf_available, is_torch_available
__a :List[Any] = {
'configuration_tapas': ['TAPAS_PRETRAINED_CONFIG_ARCHIVE_MAP', 'TapasConfig'],
'tokenization_tapas': ['TapasTokenizer'],
}
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Any = [
'TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST',
'TapasForMaskedLM',
'TapasForQuestionAnswering',
'TapasForSequenceClassification',
'TapasModel',
'TapasPreTrainedModel',
'load_tf_weights_in_tapas',
]
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Dict = [
'TF_TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST',
'TFTapasForMaskedLM',
'TFTapasForQuestionAnswering',
'TFTapasForSequenceClassification',
'TFTapasModel',
'TFTapasPreTrainedModel',
]
if TYPE_CHECKING:
from .configuration_tapas import TAPAS_PRETRAINED_CONFIG_ARCHIVE_MAP, TapasConfig
from .tokenization_tapas import TapasTokenizer
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tapas import (
TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST,
TapasForMaskedLM,
TapasForQuestionAnswering,
TapasForSequenceClassification,
TapasModel,
TapasPreTrainedModel,
load_tf_weights_in_tapas,
)
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tf_tapas import (
TF_TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST,
TFTapasForMaskedLM,
TFTapasForQuestionAnswering,
TFTapasForSequenceClassification,
TFTapasModel,
TFTapasPreTrainedModel,
)
else:
import sys
__a :str = _LazyModule(__name__, globals()['__file__'], _import_structure, module_spec=__spec__) | 86 | 0 |
import unittest
from transformers import is_torch_available
from transformers.testing_utils import require_sentencepiece, require_tokenizers, require_torch, slow
if is_torch_available():
import torch
from transformers import XLMRobertaModel
@require_sentencepiece
@require_tokenizers
@require_torch
class lowerCamelCase_ ( unittest.TestCase ):
@slow
def lowerCAmelCase_ ( self : Optional[Any] ):
__A : Union[str, Any] = XLMRobertaModel.from_pretrained("""xlm-roberta-base""" )
__A : Optional[Any] = torch.tensor([[0, 581, 1_0269, 83, 9_9942, 136, 6_0742, 23, 70, 8_0583, 1_8276, 2]] )
# The dog is cute and lives in the garden house
__A : List[str] = torch.Size((1, 12, 768) ) # batch_size, sequence_length, embedding_vector_dim
__A : int = torch.tensor(
[[-0.0_1_0_1, 0.1_2_1_8, -0.0_8_0_3, 0.0_8_0_1, 0.1_3_2_7, 0.0_7_7_6, -0.1_2_1_5, 0.2_3_8_3, 0.3_3_3_8, 0.3_1_0_6, 0.0_3_0_0, 0.0_2_5_2]] )
# xlmr = torch.hub.load('pytorch/fairseq', 'xlmr.base')
# xlmr.eval()
# expected_output_values_last_dim = xlmr.extract_features(input_ids[0])[:, :, -1]
with torch.no_grad():
__A : Dict = model(__A )["""last_hidden_state"""].detach()
self.assertEqual(output.shape , __A )
# compare the actual values for a slice of last dim
self.assertTrue(torch.allclose(output[:, :, -1] , __A , atol=1e-3 ) )
@slow
def lowerCAmelCase_ ( self : List[str] ):
__A : Any = XLMRobertaModel.from_pretrained("""xlm-roberta-large""" )
__A : int = torch.tensor([[0, 581, 1_0269, 83, 9_9942, 136, 6_0742, 23, 70, 8_0583, 1_8276, 2]] )
# The dog is cute and lives in the garden house
__A : Optional[Any] = torch.Size((1, 12, 1024) ) # batch_size, sequence_length, embedding_vector_dim
__A : Dict = torch.tensor(
[[-0.0_6_9_9, -0.0_3_1_8, 0.0_7_0_5, -0.1_2_4_1, 0.0_9_9_9, -0.0_5_2_0, 0.1_0_0_4, -0.1_8_3_8, -0.4_7_0_4, 0.1_4_3_7, 0.0_8_2_1, 0.0_1_2_6]] )
# xlmr = torch.hub.load('pytorch/fairseq', 'xlmr.large')
# xlmr.eval()
# expected_output_values_last_dim = xlmr.extract_features(input_ids[0])[:, :, -1]
with torch.no_grad():
__A : Dict = model(__A )["""last_hidden_state"""].detach()
self.assertEqual(output.shape , __A )
# compare the actual values for a slice of last dim
self.assertTrue(torch.allclose(output[:, :, -1] , __A , atol=1e-3 ) )
| 17 |
# Lint as: python3
import sys
from collections.abc import Mapping
from typing import TYPE_CHECKING, Dict, Optional
import numpy as np
import pyarrow as pa
from .. import config
from ..utils.logging import get_logger
from ..utils.py_utils import map_nested
from .formatting import TensorFormatter
if TYPE_CHECKING:
import jax
import jaxlib
__a :List[Any] = get_logger()
__a :Optional[dict] = None
class _a ( TensorFormatter[Mapping, 'jax.Array', Mapping] ):
"""simple docstring"""
def __init__( self : str , UpperCAmelCase : int=None , UpperCAmelCase : List[str]=None , **UpperCAmelCase : List[Any] ):
super().__init__(features=UpperCAmelCase )
import jax
from jaxlib.xla_client import Device
if isinstance(UpperCAmelCase , UpperCAmelCase ):
raise ValueError(
f'''Expected {device} to be a `str` not {type(UpperCAmelCase )}, as `jaxlib.xla_extension.Device` '''
"is not serializable neither with `pickle` nor with `dill`. Instead you can surround "
"the device with `str()` to get its string identifier that will be internally mapped "
"to the actual `jaxlib.xla_extension.Device`." )
A_ = device if isinstance(UpperCAmelCase , UpperCAmelCase ) else str(jax.devices()[0] )
# using global variable since `jaxlib.xla_extension.Device` is not serializable neither
# with `pickle` nor with `dill`, so we need to use a global variable instead
global DEVICE_MAPPING
if DEVICE_MAPPING is None:
A_ = self._map_devices_to_str()
if self.device not in list(DEVICE_MAPPING.keys() ):
logger.warning(
f'''Device with string identifier {self.device} not listed among the available '''
f'''devices: {list(DEVICE_MAPPING.keys() )}, so falling back to the default '''
f'''device: {str(jax.devices()[0] )}.''' )
A_ = str(jax.devices()[0] )
A_ = jnp_array_kwargs
@staticmethod
def __A ( ):
import jax
return {str(UpperCAmelCase ): device for device in jax.devices()}
def __A ( self : Optional[int] , UpperCAmelCase : int ):
import jax
import jax.numpy as jnp
if isinstance(UpperCAmelCase , UpperCAmelCase ) and column:
if all(
isinstance(UpperCAmelCase , jax.Array ) and x.shape == column[0].shape and x.dtype == column[0].dtype for x in column ):
return jnp.stack(UpperCAmelCase , axis=0 )
return column
def __A ( self : List[str] , UpperCAmelCase : str ):
import jax
import jax.numpy as jnp
if isinstance(UpperCAmelCase , (str, bytes, type(UpperCAmelCase )) ):
return value
elif isinstance(UpperCAmelCase , (np.character, np.ndarray) ) and np.issubdtype(value.dtype , np.character ):
return value.tolist()
A_ = {}
if isinstance(UpperCAmelCase , (np.number, np.ndarray) ) and np.issubdtype(value.dtype , np.integer ):
# the default int precision depends on the jax config
# see https://jax.readthedocs.io/en/latest/notebooks/Common_Gotchas_in_JAX.html#double-64bit-precision
if jax.config.jax_enable_xaa:
A_ = {"dtype": jnp.intaa}
else:
A_ = {"dtype": jnp.intaa}
elif isinstance(UpperCAmelCase , (np.number, np.ndarray) ) and np.issubdtype(value.dtype , np.floating ):
A_ = {"dtype": jnp.floataa}
elif config.PIL_AVAILABLE and "PIL" in sys.modules:
import PIL.Image
if isinstance(UpperCAmelCase , PIL.Image.Image ):
A_ = np.asarray(UpperCAmelCase )
# using global variable since `jaxlib.xla_extension.Device` is not serializable neither
# with `pickle` nor with `dill`, so we need to use a global variable instead
global DEVICE_MAPPING
if DEVICE_MAPPING is None:
A_ = self._map_devices_to_str()
with jax.default_device(DEVICE_MAPPING[self.device] ):
# calling jnp.array on a np.ndarray does copy the data
# see https://github.com/google/jax/issues/4486
return jnp.array(UpperCAmelCase , **{**default_dtype, **self.jnp_array_kwargs} )
def __A ( self : Any , UpperCAmelCase : Dict ):
import jax
# support for torch, tf, jax etc.
if config.TORCH_AVAILABLE and "torch" in sys.modules:
import torch
if isinstance(UpperCAmelCase , torch.Tensor ):
return self._tensorize(data_struct.detach().cpu().numpy()[()] )
if hasattr(UpperCAmelCase , "__array__" ) and not isinstance(UpperCAmelCase , jax.Array ):
A_ = data_struct.__array__()
# support for nested types like struct of list of struct
if isinstance(UpperCAmelCase , np.ndarray ):
if data_struct.dtype == object: # jax arrays cannot be instantied from an array of objects
return self._consolidate([self.recursive_tensorize(UpperCAmelCase ) for substruct in data_struct] )
elif isinstance(UpperCAmelCase , (list, tuple) ):
return self._consolidate([self.recursive_tensorize(UpperCAmelCase ) for substruct in data_struct] )
return self._tensorize(UpperCAmelCase )
def __A ( self : Tuple , UpperCAmelCase : dict ):
return map_nested(self._recursive_tensorize , UpperCAmelCase , map_list=UpperCAmelCase )
def __A ( self : Dict , UpperCAmelCase : pa.Table ):
A_ = self.numpy_arrow_extractor().extract_row(UpperCAmelCase )
A_ = self.python_features_decoder.decode_row(UpperCAmelCase )
return self.recursive_tensorize(UpperCAmelCase )
def __A ( self : Any , UpperCAmelCase : pa.Table ):
A_ = self.numpy_arrow_extractor().extract_column(UpperCAmelCase )
A_ = self.python_features_decoder.decode_column(UpperCAmelCase , pa_table.column_names[0] )
A_ = self.recursive_tensorize(UpperCAmelCase )
A_ = self._consolidate(UpperCAmelCase )
return column
def __A ( self : Dict , UpperCAmelCase : pa.Table ):
A_ = self.numpy_arrow_extractor().extract_batch(UpperCAmelCase )
A_ = self.python_features_decoder.decode_batch(UpperCAmelCase )
A_ = self.recursive_tensorize(UpperCAmelCase )
for column_name in batch:
A_ = self._consolidate(batch[column_name] )
return batch | 86 | 0 |
'''simple docstring'''
from ...configuration_utils import PretrainedConfig
from ...utils import logging
_SCREAMING_SNAKE_CASE = logging.get_logger(__name__)
_SCREAMING_SNAKE_CASE = {
"microsoft/swinv2-tiny-patch4-window8-256": (
"https://huggingface.co/microsoft/swinv2-tiny-patch4-window8-256/resolve/main/config.json"
),
}
class lowerCAmelCase_ ( __magic_name__ ):
__lowerCamelCase : Union[str, Any] = "swinv2"
__lowerCamelCase : int = {
"num_attention_heads": "num_heads",
"num_hidden_layers": "num_layers",
}
def __init__( self , _lowerCAmelCase=224 , _lowerCAmelCase=4 , _lowerCAmelCase=3 , _lowerCAmelCase=96 , _lowerCAmelCase=[2, 2, 6, 2] , _lowerCAmelCase=[3, 6, 12, 24] , _lowerCAmelCase=7 , _lowerCAmelCase=4.0 , _lowerCAmelCase=True , _lowerCAmelCase=0.0 , _lowerCAmelCase=0.0 , _lowerCAmelCase=0.1 , _lowerCAmelCase="gelu" , _lowerCAmelCase=False , _lowerCAmelCase=0.02 , _lowerCAmelCase=1E-5 , _lowerCAmelCase=32 , **_lowerCAmelCase , ) -> Tuple:
super().__init__(**_lowerCAmelCase )
_lowerCAmelCase = image_size
_lowerCAmelCase = patch_size
_lowerCAmelCase = num_channels
_lowerCAmelCase = embed_dim
_lowerCAmelCase = depths
_lowerCAmelCase = len(_lowerCAmelCase )
_lowerCAmelCase = num_heads
_lowerCAmelCase = window_size
_lowerCAmelCase = mlp_ratio
_lowerCAmelCase = qkv_bias
_lowerCAmelCase = hidden_dropout_prob
_lowerCAmelCase = attention_probs_dropout_prob
_lowerCAmelCase = drop_path_rate
_lowerCAmelCase = hidden_act
_lowerCAmelCase = use_absolute_embeddings
_lowerCAmelCase = layer_norm_eps
_lowerCAmelCase = initializer_range
_lowerCAmelCase = encoder_stride
# we set the hidden_size attribute in order to make Swinv2 work with VisionEncoderDecoderModel
# this indicates the channel dimension after the last stage of the model
_lowerCAmelCase = int(embed_dim * 2 ** (len(_lowerCAmelCase ) - 1) )
_lowerCAmelCase = (0, 0, 0, 0)
| 18 |
import logging
import os
from typing import List, Tuple
import numpy as np
import psutil
import torch
import torch.distributed as dist
from transformers import RagRetriever
__a :Any = logging.getLogger(__name__)
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Dict , UpperCAmelCase : List[Any] , UpperCAmelCase : Optional[int] , UpperCAmelCase : Optional[Any] , UpperCAmelCase : str=None ):
super().__init__(
UpperCAmelCase , question_encoder_tokenizer=UpperCAmelCase , generator_tokenizer=UpperCAmelCase , index=UpperCAmelCase , init_retrieval=UpperCAmelCase , )
A_ = None
def __A ( self : Dict , UpperCAmelCase : int ):
logger.info("initializing retrieval" )
# initializing a separate process group for retrieval as the default
# nccl backend doesn't support gather/scatter operations while gloo
# is too slow to replace nccl for the core gpu communication
if dist.is_initialized():
logger.info("dist initialized" )
# needs to be set manually
A_ = self._infer_socket_ifname()
# avoid clash with the NCCL port
A_ = str(distributed_port + 1 )
A_ = dist.new_group(ranks=UpperCAmelCase , backend="gloo" )
# initialize retriever only on the main worker
if not dist.is_initialized() or self._is_main():
logger.info("dist not initialized / main" )
self.index.init_index()
# all processes wait untill the retriever is initialized by the main process
if dist.is_initialized():
torch.distributed.barrier(group=self.process_group )
def __A ( self : List[str] ):
return dist.get_rank(group=self.process_group ) == 0
def __A ( self : Any , UpperCAmelCase : List[Any] , UpperCAmelCase : Union[str, Any] , UpperCAmelCase : Dict=torch.floataa ):
A_ = torch.empty(UpperCAmelCase , dtype=UpperCAmelCase )
dist.scatter(UpperCAmelCase , src=0 , scatter_list=UpperCAmelCase , group=self.process_group )
return target_tensor
def __A ( self : Any ):
A_ = psutil.net_if_addrs()
# a hacky way to deal with varying network interface names
A_ = next((addr for addr in addrs if addr.startswith("e" )) , UpperCAmelCase )
return ifname
def __A ( self : Tuple , UpperCAmelCase : np.ndarray , UpperCAmelCase : int ):
# single GPU training
if not dist.is_initialized():
A_ , A_ = self._main_retrieve(UpperCAmelCase , UpperCAmelCase )
return retrieved_doc_embeds, doc_ids, self.index.get_doc_dicts(UpperCAmelCase )
# distributed training
A_ = dist.get_world_size(group=self.process_group )
# gather logic
A_ = None
if self._is_main():
A_ = [torch.empty(question_hidden_states.shape , dtype=torch.floataa ) for _ in range(UpperCAmelCase )]
dist.gather(torch.tensor(UpperCAmelCase ) , dst=0 , gather_list=UpperCAmelCase , group=self.process_group )
# scatter logic
A_ = question_hidden_states.shape[0]
A_ = []
A_ = []
if self._is_main():
assert len(UpperCAmelCase ) == world_size
A_ , A_ = self._main_retrieve(torch.cat(UpperCAmelCase ).numpy() , UpperCAmelCase )
A_ , A_ = torch.tensor(UpperCAmelCase ), torch.tensor(UpperCAmelCase )
A_ = self._chunk_tensor(UpperCAmelCase , UpperCAmelCase )
A_ = self._chunk_tensor(UpperCAmelCase , UpperCAmelCase )
A_ = self._scattered(UpperCAmelCase , [n_queries, n_docs] , target_type=torch.intaa )
A_ = self._scattered(UpperCAmelCase , [n_queries, n_docs, question_hidden_states.shape[1]] )
return retrieved_doc_embeds.numpy(), doc_ids.numpy(), self.index.get_doc_dicts(UpperCAmelCase ) | 86 | 0 |
"""simple docstring"""
from timeit import timeit
_a = {
"""MALAYALAM""": True,
"""String""": False,
"""rotor""": True,
"""level""": True,
"""A""": True,
"""BB""": True,
"""ABC""": False,
"""amanaplanacanalpanama""": True, # "a man a plan a canal panama"
}
# Ensure our test data is valid
assert all((key == key[::-1]) is value for key, value in test_data.items())
def lowerCamelCase__ ( __snake_case ) -> bool:
"""simple docstring"""
_UpperCamelCase = 0
_UpperCamelCase = len(__snake_case ) - 1
while start_i < end_i:
if s[start_i] == s[end_i]:
start_i += 1
end_i -= 1
else:
return False
return True
def lowerCamelCase__ ( __snake_case ) -> bool:
"""simple docstring"""
_UpperCamelCase = len(__snake_case ) // 2
_UpperCamelCase = len(__snake_case )
# We need to traverse till half of the length of string
# as we can get access of the i'th last element from
# i'th index.
# eg: [0,1,2,3,4,5] => 4th index can be accessed
# with the help of 1st index (i==n-i-1)
# where n is length of string
return all(s[i] == s[n - i - 1] for i in range(__snake_case ) )
def lowerCamelCase__ ( __snake_case ) -> bool:
"""simple docstring"""
if len(__snake_case ) <= 2:
return True
if s[0] == s[len(__snake_case ) - 1]:
return is_palindrome_recursive(s[1:-1] )
else:
return False
def lowerCamelCase__ ( __snake_case ) -> bool:
"""simple docstring"""
return s == s[::-1]
def lowerCamelCase__ ( __snake_case ) -> None:
"""simple docstring"""
_UpperCamelCase = F'''all({name}(key) is value for key, value in test_data.items())'''
_UpperCamelCase = F'''from __main__ import test_data, {name}'''
_UpperCamelCase = 50_00_00
_UpperCamelCase = timeit(stmt=__snake_case, setup=__snake_case, number=__snake_case )
print(F'''{name:<35} finished {number:,} runs in {result:.5f} seconds''' )
if __name__ == "__main__":
for key, value in test_data.items():
assert is_palindrome(key) is is_palindrome_recursive(key)
assert is_palindrome(key) is is_palindrome_slice(key)
print(F"""{key:21} {value}""")
print("""a man a plan a canal panama""")
# finished 500,000 runs in 0.46793 seconds
benchmark_function("""is_palindrome_slice""")
# finished 500,000 runs in 0.85234 seconds
benchmark_function("""is_palindrome""")
# finished 500,000 runs in 1.32028 seconds
benchmark_function("""is_palindrome_recursive""")
# finished 500,000 runs in 2.08679 seconds
benchmark_function("""is_palindrome_traversal""")
| 19 |
from jiwer import compute_measures
import datasets
__a :List[Any] = '\\n@inproceedings{inproceedings,\n author = {Morris, Andrew and Maier, Viktoria and Green, Phil},\n year = {2004},\n month = {01},\n pages = {},\n title = {From WER and RIL to MER and WIL: improved evaluation measures for connected speech recognition.}\n}\n'
__a :Union[str, Any] = '\\nWord error rate (WER) is a common metric of the performance of an automatic speech recognition system.\n\nThe general difficulty of measuring performance lies in the fact that the recognized word sequence can have a different length from the reference word sequence (supposedly the correct one). The WER is derived from the Levenshtein distance, working at the word level instead of the phoneme level. The WER is a valuable tool for comparing different systems as well as for evaluating improvements within one system. This kind of measurement, however, provides no details on the nature of translation errors and further work is therefore required to identify the main source(s) of error and to focus any research effort.\n\nThis problem is solved by first aligning the recognized word sequence with the reference (spoken) word sequence using dynamic string alignment. Examination of this issue is seen through a theory called the power law that states the correlation between perplexity and word error rate.\n\nWord error rate can then be computed as:\n\nWER = (S + D + I) / N = (S + D + I) / (S + D + C)\n\nwhere\n\nS is the number of substitutions,\nD is the number of deletions,\nI is the number of insertions,\nC is the number of correct words,\nN is the number of words in the reference (N=S+D+C).\n\nThis value indicates the average number of errors per reference word. The lower the value, the better the\nperformance of the ASR system with a WER of 0 being a perfect score.\n'
__a :str = '\nCompute WER score of transcribed segments against references.\n\nArgs:\n references: List of references for each speech input.\n predictions: List of transcriptions to score.\n concatenate_texts (bool, default=False): Whether to concatenate all input texts or compute WER iteratively.\n\nReturns:\n (float): the word error rate\n\nExamples:\n\n >>> predictions = ["this is the prediction", "there is an other sample"]\n >>> references = ["this is the reference", "there is another one"]\n >>> wer = datasets.load_metric("wer")\n >>> wer_score = wer.compute(predictions=predictions, references=references)\n >>> print(wer_score)\n 0.5\n'
@datasets.utils.file_utils.add_start_docstrings(_DESCRIPTION , _KWARGS_DESCRIPTION )
class _a ( datasets.Metric ):
"""simple docstring"""
def __A ( self : Any ):
return datasets.MetricInfo(
description=_DESCRIPTION , citation=_CITATION , inputs_description=_KWARGS_DESCRIPTION , features=datasets.Features(
{
"predictions": datasets.Value("string" , id="sequence" ),
"references": datasets.Value("string" , id="sequence" ),
} ) , codebase_urls=["https://github.com/jitsi/jiwer/"] , reference_urls=[
"https://en.wikipedia.org/wiki/Word_error_rate",
] , )
def __A ( self : Dict , UpperCAmelCase : Dict=None , UpperCAmelCase : List[Any]=None , UpperCAmelCase : str=False ):
if concatenate_texts:
return compute_measures(UpperCAmelCase , UpperCAmelCase )["wer"]
else:
A_ = 0
A_ = 0
for prediction, reference in zip(UpperCAmelCase , UpperCAmelCase ):
A_ = compute_measures(UpperCAmelCase , UpperCAmelCase )
incorrect += measures["substitutions"] + measures["deletions"] + measures["insertions"]
total += measures["substitutions"] + measures["deletions"] + measures["hits"]
return incorrect / total | 86 | 0 |
def _lowercase( __a : list ):
if not isinstance(__a , __a ):
raise ValueError('Input series is not valid, valid series - [2, 4, 6]' )
if len(__a ) == 0:
raise ValueError('Input list must be a non empty list' )
if len(__a ) == 1:
return True
a__ =series[1] - series[0]
for index in range(len(__a ) - 1 ):
if series[index + 1] - series[index] != common_diff:
return False
return True
def _lowercase( __a : list ):
if not isinstance(__a , __a ):
raise ValueError('Input series is not valid, valid series - [2, 4, 6]' )
if len(__a ) == 0:
raise ValueError('Input list must be a non empty list' )
a__ =0
for val in series:
answer += val
return answer / len(__a )
if __name__ == "__main__":
import doctest
doctest.testmod()
| 20 |
class _a :
"""simple docstring"""
def __init__( self : Union[str, Any] , UpperCAmelCase : int , UpperCAmelCase : Any , UpperCAmelCase : Dict ):
A_ = None
A_ = None
A_ = graph
self._normalize_graph(UpperCAmelCase , UpperCAmelCase )
A_ = len(UpperCAmelCase )
A_ = None
def __A ( self : Union[str, Any] , UpperCAmelCase : List[str] , UpperCAmelCase : Tuple ):
if sources is int:
A_ = [sources]
if sinks is int:
A_ = [sinks]
if len(UpperCAmelCase ) == 0 or len(UpperCAmelCase ) == 0:
return
A_ = sources[0]
A_ = sinks[0]
# make fake vertex if there are more
# than one source or sink
if len(UpperCAmelCase ) > 1 or len(UpperCAmelCase ) > 1:
A_ = 0
for i in sources:
max_input_flow += sum(self.graph[i] )
A_ = len(self.graph ) + 1
for room in self.graph:
room.insert(0 , 0 )
self.graph.insert(0 , [0] * size )
for i in sources:
A_ = max_input_flow
A_ = 0
A_ = len(self.graph ) + 1
for room in self.graph:
room.append(0 )
self.graph.append([0] * size )
for i in sinks:
A_ = max_input_flow
A_ = size - 1
def __A ( self : str ):
if self.maximum_flow_algorithm is None:
raise Exception("You need to set maximum flow algorithm before." )
if self.source_index is None or self.sink_index is None:
return 0
self.maximum_flow_algorithm.execute()
return self.maximum_flow_algorithm.getMaximumFlow()
def __A ( self : Tuple , UpperCAmelCase : List[Any] ):
A_ = algorithm(self )
class _a :
"""simple docstring"""
def __init__( self : Tuple , UpperCAmelCase : List[str] ):
A_ = flow_network
A_ = flow_network.verticesCount
A_ = flow_network.sourceIndex
A_ = flow_network.sinkIndex
# it's just a reference, so you shouldn't change
# it in your algorithms, use deep copy before doing that
A_ = flow_network.graph
A_ = False
def __A ( self : Optional[int] ):
if not self.executed:
self._algorithm()
A_ = True
def __A ( self : Dict ):
pass
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Optional[Any] , UpperCAmelCase : List[Any] ):
super().__init__(UpperCAmelCase )
# use this to save your result
A_ = -1
def __A ( self : Tuple ):
if not self.executed:
raise Exception("You should execute algorithm before using its result!" )
return self.maximum_flow
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Tuple , UpperCAmelCase : Union[str, Any] ):
super().__init__(UpperCAmelCase )
A_ = [[0] * self.verticies_count for i in range(self.verticies_count )]
A_ = [0] * self.verticies_count
A_ = [0] * self.verticies_count
def __A ( self : List[str] ):
A_ = self.verticies_count
# push some substance to graph
for nextvertex_index, bandwidth in enumerate(self.graph[self.source_index] ):
self.preflow[self.source_index][nextvertex_index] += bandwidth
self.preflow[nextvertex_index][self.source_index] -= bandwidth
self.excesses[nextvertex_index] += bandwidth
# Relabel-to-front selection rule
A_ = [
i
for i in range(self.verticies_count )
if i != self.source_index and i != self.sink_index
]
# move through list
A_ = 0
while i < len(UpperCAmelCase ):
A_ = vertices_list[i]
A_ = self.heights[vertex_index]
self.process_vertex(UpperCAmelCase )
if self.heights[vertex_index] > previous_height:
# if it was relabeled, swap elements
# and start from 0 index
vertices_list.insert(0 , vertices_list.pop(UpperCAmelCase ) )
A_ = 0
else:
i += 1
A_ = sum(self.preflow[self.source_index] )
def __A ( self : List[str] , UpperCAmelCase : Dict ):
while self.excesses[vertex_index] > 0:
for neighbour_index in range(self.verticies_count ):
# if it's neighbour and current vertex is higher
if (
self.graph[vertex_index][neighbour_index]
- self.preflow[vertex_index][neighbour_index]
> 0
and self.heights[vertex_index] > self.heights[neighbour_index]
):
self.push(UpperCAmelCase , UpperCAmelCase )
self.relabel(UpperCAmelCase )
def __A ( self : Optional[Any] , UpperCAmelCase : Dict , UpperCAmelCase : str ):
A_ = min(
self.excesses[from_index] , self.graph[from_index][to_index] - self.preflow[from_index][to_index] , )
self.preflow[from_index][to_index] += preflow_delta
self.preflow[to_index][from_index] -= preflow_delta
self.excesses[from_index] -= preflow_delta
self.excesses[to_index] += preflow_delta
def __A ( self : Optional[Any] , UpperCAmelCase : List[Any] ):
A_ = None
for to_index in range(self.verticies_count ):
if (
self.graph[vertex_index][to_index]
- self.preflow[vertex_index][to_index]
> 0
) and (min_height is None or self.heights[to_index] < min_height):
A_ = self.heights[to_index]
if min_height is not None:
A_ = min_height + 1
if __name__ == "__main__":
__a :Tuple = [0]
__a :Tuple = [3]
# graph = [
# [0, 0, 4, 6, 0, 0],
# [0, 0, 5, 2, 0, 0],
# [0, 0, 0, 0, 4, 4],
# [0, 0, 0, 0, 6, 6],
# [0, 0, 0, 0, 0, 0],
# [0, 0, 0, 0, 0, 0],
# ]
__a :List[str] = [[0, 7, 0, 0], [0, 0, 6, 0], [0, 0, 0, 8], [9, 0, 0, 0]]
# prepare our network
__a :List[str] = FlowNetwork(graph, entrances, exits)
# set algorithm
flow_network.set_maximum_flow_algorithm(PushRelabelExecutor)
# and calculate
__a :List[Any] = flow_network.find_maximum_flow()
print(F"maximum flow is {maximum_flow}") | 86 | 0 |
from __future__ import annotations
def lowerCAmelCase_ ( lowerCamelCase , lowerCamelCase , lowerCamelCase , lowerCamelCase ):
__magic_name__ : str =[]
__magic_name__ , __magic_name__ : str =input_list[low:mid], input_list[mid : high + 1]
while left and right:
result.append((left if left[0] <= right[0] else right).pop(0 ) )
__magic_name__ : int =result + left + right
return input_list
def lowerCAmelCase_ ( lowerCamelCase ):
if len(lowerCamelCase ) <= 1:
return input_list
__magic_name__ : Any =list(lowerCamelCase )
# iteration for two-way merging
__magic_name__ : Optional[Any] =2
while p <= len(lowerCamelCase ):
# getting low, high and middle value for merge-sort of single list
for i in range(0 , len(lowerCamelCase ) , lowerCamelCase ):
__magic_name__ : Union[str, Any] =i
__magic_name__ : Union[str, Any] =i + p - 1
__magic_name__ : Dict =(low + high + 1) // 2
__magic_name__ : str =merge(lowerCamelCase , lowerCamelCase , lowerCamelCase , lowerCamelCase )
# final merge of last two parts
if p * 2 >= len(lowerCamelCase ):
__magic_name__ : Any =i
__magic_name__ : Any =merge(lowerCamelCase , 0 , lowerCamelCase , len(lowerCamelCase ) - 1 )
break
p *= 2
return input_list
if __name__ == "__main__":
UpperCAmelCase_ : Dict = input("Enter numbers separated by a comma:\n").strip()
if user_input == "":
UpperCAmelCase_ : Optional[Any] = []
else:
UpperCAmelCase_ : Optional[int] = [int(item.strip()) for item in user_input.split(",")]
print(iter_merge_sort(unsorted))
| 21 |
from typing import TYPE_CHECKING
from ...utils import (
OptionalDependencyNotAvailable,
_LazyModule,
is_flax_available,
is_sentencepiece_available,
is_tf_available,
is_tokenizers_available,
is_torch_available,
)
__a :Dict = {'configuration_xglm': ['XGLM_PRETRAINED_CONFIG_ARCHIVE_MAP', 'XGLMConfig']}
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Dict = ['XGLMTokenizer']
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :str = ['XGLMTokenizerFast']
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Tuple = [
'XGLM_PRETRAINED_MODEL_ARCHIVE_LIST',
'XGLMForCausalLM',
'XGLMModel',
'XGLMPreTrainedModel',
]
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :List[Any] = [
'FlaxXGLMForCausalLM',
'FlaxXGLMModel',
'FlaxXGLMPreTrainedModel',
]
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Any = [
'TF_XGLM_PRETRAINED_MODEL_ARCHIVE_LIST',
'TFXGLMForCausalLM',
'TFXGLMModel',
'TFXGLMPreTrainedModel',
]
if TYPE_CHECKING:
from .configuration_xglm import XGLM_PRETRAINED_CONFIG_ARCHIVE_MAP, XGLMConfig
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_xglm import XGLMTokenizer
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_xglm_fast import XGLMTokenizerFast
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_xglm import XGLM_PRETRAINED_MODEL_ARCHIVE_LIST, XGLMForCausalLM, XGLMModel, XGLMPreTrainedModel
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_flax_xglm import FlaxXGLMForCausalLM, FlaxXGLMModel, FlaxXGLMPreTrainedModel
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tf_xglm import (
TF_XGLM_PRETRAINED_MODEL_ARCHIVE_LIST,
TFXGLMForCausalLM,
TFXGLMModel,
TFXGLMPreTrainedModel,
)
else:
import sys
__a :List[Any] = _LazyModule(__name__, globals()['__file__'], _import_structure) | 86 | 0 |
'''simple docstring'''
from ...configuration_utils import PretrainedConfig
from ...utils import logging
_snake_case : Optional[Any] = logging.get_logger(__name__)
_snake_case : str = {
'tanreinama/GPTSAN-2.8B-spout_is_uniform': (
'https://huggingface.co/tanreinama/GPTSAN-2.8B-spout_is_uniform/resolve/main/config.json'
),
}
class A ( _a ):
lowercase_ = 'gptsan-japanese'
lowercase_ = [
'past_key_values',
]
lowercase_ = {
'hidden_size': 'd_model',
'num_attention_heads': 'num_heads',
'num_hidden_layers': 'num_layers',
}
def __init__( self : Union[str, Any] , lowerCAmelCase_ : Dict=3_60_00 , lowerCAmelCase_ : Union[str, Any]=12_80 , lowerCAmelCase_ : Optional[int]=10_24 , lowerCAmelCase_ : Union[str, Any]=81_92 , lowerCAmelCase_ : int=40_96 , lowerCAmelCase_ : Dict=1_28 , lowerCAmelCase_ : Optional[int]=10 , lowerCAmelCase_ : Union[str, Any]=0 , lowerCAmelCase_ : Optional[Any]=16 , lowerCAmelCase_ : Optional[int]=16 , lowerCAmelCase_ : str=1_28 , lowerCAmelCase_ : Optional[int]=0.0 , lowerCAmelCase_ : int=1e-5 , lowerCAmelCase_ : Union[str, Any]=False , lowerCAmelCase_ : List[str]=0.0 , lowerCAmelCase_ : Union[str, Any]="float32" , lowerCAmelCase_ : Optional[Any]=False , lowerCAmelCase_ : Any=False , lowerCAmelCase_ : List[Any]=False , lowerCAmelCase_ : List[str]=0.0_0_2 , lowerCAmelCase_ : Dict=False , lowerCAmelCase_ : List[Any]=True , lowerCAmelCase_ : List[Any]=3_59_98 , lowerCAmelCase_ : Tuple=3_59_95 , lowerCAmelCase_ : Optional[Any]=3_59_99 , **lowerCAmelCase_ : Any , ) -> List[str]:
"""simple docstring"""
_a = vocab_size
_a = max_position_embeddings
_a = d_model
_a = d_ff
_a = d_ext
_a = d_spout
_a = num_switch_layers
_a = num_ext_layers
_a = num_switch_layers + num_ext_layers
_a = num_heads
_a = num_experts
_a = expert_capacity
_a = dropout_rate
_a = layer_norm_epsilon
_a = router_bias
_a = router_jitter_noise
_a = router_dtype
_a = router_ignore_padding_tokens
_a = output_hidden_states
_a = output_attentions
_a = initializer_factor
_a = output_router_logits
_a = use_cache
super().__init__(
separator_token_id=lowerCAmelCase_ , pad_token_id=lowerCAmelCase_ , eos_token_id=lowerCAmelCase_ , **lowerCAmelCase_ , )
| 22 |
# Usage:
# ./gen-card-facebook-wmt19.py
import os
from pathlib import Path
def __snake_case ( __UpperCamelCase : str ,__UpperCamelCase : str ,__UpperCamelCase : List[str] ):
"""simple docstring"""
A_ = {
"en": "Machine learning is great, isn't it?",
"ru": "Машинное обучение - это здорово, не так ли?",
"de": "Maschinelles Lernen ist großartig, oder?",
}
# BLUE scores as follows:
# "pair": [fairseq, transformers]
A_ = {
"ru-en": ["[41.3](http://matrix.statmt.org/matrix/output/1907?run_id=6937)", "39.20"],
"en-ru": ["[36.4](http://matrix.statmt.org/matrix/output/1914?run_id=6724)", "33.47"],
"en-de": ["[43.1](http://matrix.statmt.org/matrix/output/1909?run_id=6862)", "42.83"],
"de-en": ["[42.3](http://matrix.statmt.org/matrix/output/1902?run_id=6750)", "41.35"],
}
A_ = f'''{src_lang}-{tgt_lang}'''
A_ = f'''
---
language:
- {src_lang}
- {tgt_lang}
thumbnail:
tags:
- translation
- wmt19
- facebook
license: apache-2.0
datasets:
- wmt19
metrics:
- bleu
---
# FSMT
## Model description
This is a ported version of [fairseq wmt19 transformer](https://github.com/pytorch/fairseq/blob/master/examples/wmt19/README.md) for {src_lang}-{tgt_lang}.
For more details, please see, [Facebook FAIR\'s WMT19 News Translation Task Submission](https://arxiv.org/abs/1907.06616).
The abbreviation FSMT stands for FairSeqMachineTranslation
All four models are available:
* [wmt19-en-ru](https://huggingface.co/facebook/wmt19-en-ru)
* [wmt19-ru-en](https://huggingface.co/facebook/wmt19-ru-en)
* [wmt19-en-de](https://huggingface.co/facebook/wmt19-en-de)
* [wmt19-de-en](https://huggingface.co/facebook/wmt19-de-en)
## Intended uses & limitations
#### How to use
```python
from transformers import FSMTForConditionalGeneration, FSMTTokenizer
mname = "facebook/wmt19-{src_lang}-{tgt_lang}"
tokenizer = FSMTTokenizer.from_pretrained(mname)
model = FSMTForConditionalGeneration.from_pretrained(mname)
input = "{texts[src_lang]}"
input_ids = tokenizer.encode(input, return_tensors="pt")
outputs = model.generate(input_ids)
decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(decoded) # {texts[tgt_lang]}
```
#### Limitations and bias
- The original (and this ported model) doesn\'t seem to handle well inputs with repeated sub-phrases, [content gets truncated](https://discuss.huggingface.co/t/issues-with-translating-inputs-containing-repeated-phrases/981)
## Training data
Pretrained weights were left identical to the original model released by fairseq. For more details, please, see the [paper](https://arxiv.org/abs/1907.06616).
## Eval results
pair | fairseq | transformers
-------|---------|----------
{pair} | {scores[pair][0]} | {scores[pair][1]}
The score is slightly below the score reported by `fairseq`, since `transformers`` currently doesn\'t support:
- model ensemble, therefore the best performing checkpoint was ported (``model4.pt``).
- re-ranking
The score was calculated using this code:
```bash
git clone https://github.com/huggingface/transformers
cd transformers
export PAIR={pair}
export DATA_DIR=data/$PAIR
export SAVE_DIR=data/$PAIR
export BS=8
export NUM_BEAMS=15
mkdir -p $DATA_DIR
sacrebleu -t wmt19 -l $PAIR --echo src > $DATA_DIR/val.source
sacrebleu -t wmt19 -l $PAIR --echo ref > $DATA_DIR/val.target
echo $PAIR
PYTHONPATH="src:examples/seq2seq" python examples/seq2seq/run_eval.py facebook/wmt19-$PAIR $DATA_DIR/val.source $SAVE_DIR/test_translations.txt --reference_path $DATA_DIR/val.target --score_path $SAVE_DIR/test_bleu.json --bs $BS --task translation --num_beams $NUM_BEAMS
```
note: fairseq reports using a beam of 50, so you should get a slightly higher score if re-run with `--num_beams 50`.
## Data Sources
- [training, etc.](http://www.statmt.org/wmt19/)
- [test set](http://matrix.statmt.org/test_sets/newstest2019.tgz?1556572561)
### BibTeX entry and citation info
```bibtex
@inproceedings{{...,
year={{2020}},
title={{Facebook FAIR\'s WMT19 News Translation Task Submission}},
author={{Ng, Nathan and Yee, Kyra and Baevski, Alexei and Ott, Myle and Auli, Michael and Edunov, Sergey}},
booktitle={{Proc. of WMT}},
}}
```
## TODO
- port model ensemble (fairseq uses 4 model checkpoints)
'''
os.makedirs(__UpperCamelCase ,exist_ok=__UpperCamelCase )
A_ = os.path.join(__UpperCamelCase ,"README.md" )
print(f'''Generating {path}''' )
with open(__UpperCamelCase ,"w" ,encoding="utf-8" ) as f:
f.write(__UpperCamelCase )
# make sure we are under the root of the project
__a :Optional[Any] = Path(__file__).resolve().parent.parent.parent
__a :Optional[Any] = repo_dir / 'model_cards'
for model_name in ["wmt19-ru-en", "wmt19-en-ru", "wmt19-en-de", "wmt19-de-en"]:
__a , __a , __a :int = model_name.split('-')
__a :str = model_cards_dir / 'facebook' / model_name
write_model_card(model_card_dir, src_lang=src_lang, tgt_lang=tgt_lang) | 86 | 0 |
import gc
import importlib.metadata
import tempfile
import unittest
from packaging import version
from transformers import (
AutoModel,
AutoModelForCausalLM,
AutoModelForSeqaSeqLM,
AutoModelForSequenceClassification,
AutoTokenizer,
BitsAndBytesConfig,
pipeline,
)
from transformers.testing_utils import (
is_torch_available,
require_accelerate,
require_bitsandbytes,
require_torch,
require_torch_gpu,
require_torch_multi_gpu,
slow,
)
def _snake_case (__lowercase):
if model.config.model_type == "gpt2":
return model.transformer.h[0].mlp.c_fc
return model.transformer.h[0].mlp.dense_ah_to_h
if is_torch_available():
import torch
import torch.nn as nn
class _a ( nn.Module ):
"""simple docstring"""
def __init__( self , _UpperCAmelCase , _UpperCAmelCase ) -> List[Any]:
super().__init__()
UpperCamelCase_ = module
UpperCamelCase_ = nn.Sequential(
nn.Linear(module.in_features , _UpperCAmelCase , bias=_UpperCAmelCase ) , nn.Linear(_UpperCAmelCase , module.out_features , bias=_UpperCAmelCase ) , )
UpperCamelCase_ = (2.0 / (5 * min(module.in_features , module.out_features ))) ** 0.5
nn.init.normal_(self.adapter[0].weight , std=_UpperCAmelCase )
nn.init.zeros_(self.adapter[1].weight )
self.adapter.to(module.weight.device )
def _UpperCAmelCase ( self , _UpperCAmelCase , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[int]:
return self.module(_UpperCAmelCase , *_UpperCAmelCase , **_UpperCAmelCase ) + self.adapter(_UpperCAmelCase )
@require_bitsandbytes
@require_accelerate
@require_torch
@require_torch_gpu
@slow
class _a ( unittest.TestCase ):
"""simple docstring"""
A_ = """bigscience/bloom-1b7"""
# Constant values
A_ = 2.109_659_552_692_574
A_ = """Hello my name is"""
A_ = set()
EXPECTED_OUTPUTS.add("""Hello my name is John and I am a professional photographer. I""" )
EXPECTED_OUTPUTS.add("""Hello my name is John.\nI am a friend of your father.\n""" )
EXPECTED_OUTPUTS.add("""Hello my name is John Doe, I am a student at the University""" )
A_ = 10
def _UpperCAmelCase ( self ) -> List[Any]:
# Models and tokenizer
UpperCamelCase_ = AutoTokenizer.from_pretrained(self.model_name )
class _a ( UpperCAmelCase__ ):
"""simple docstring"""
def _UpperCAmelCase ( self ) -> List[Any]:
super().setUp()
# Models and tokenizer
UpperCamelCase_ = AutoModelForCausalLM.from_pretrained(
self.model_name , torch_dtype=torch.floataa , device_map='auto' )
UpperCamelCase_ = AutoModelForCausalLM.from_pretrained(self.model_name , load_in_abit=_UpperCAmelCase , device_map='auto' )
def _UpperCAmelCase ( self ) -> Dict:
del self.model_fpaa
del self.model_abit
gc.collect()
torch.cuda.empty_cache()
def _UpperCAmelCase ( self ) -> Tuple:
UpperCamelCase_ = self.model_abit.config
self.assertTrue(hasattr(_UpperCAmelCase , 'quantization_config' ) )
UpperCamelCase_ = config.to_dict()
UpperCamelCase_ = config.to_diff_dict()
UpperCamelCase_ = config.to_json_string()
def _UpperCAmelCase ( self ) -> int:
from bitsandbytes.nn import Paramsabit
UpperCamelCase_ = self.model_fpaa.get_memory_footprint()
UpperCamelCase_ = self.model_abit.get_memory_footprint()
self.assertAlmostEqual(mem_fpaa / mem_abit , self.EXPECTED_RELATIVE_DIFFERENCE )
UpperCamelCase_ = get_some_linear_layer(self.model_abit )
self.assertTrue(linear.weight.__class__ == Paramsabit )
def _UpperCAmelCase ( self ) -> Any:
from transformers import TaPreTrainedModel
self.model_fpaa.get_memory_footprint()
self.model_abit.get_memory_footprint()
for name, module in self.model_abit.named_modules():
if isinstance(_UpperCAmelCase , torch.nn.Linear ):
if name not in ["lm_head"] + TaPreTrainedModel._keep_in_fpaa_modules:
# 4-bit parameters are packed in uint8 variables
self.assertTrue(module.weight.dtype == torch.uinta )
def _UpperCAmelCase ( self ) -> Dict:
UpperCamelCase_ = self.tokenizer(self.input_text , return_tensors='pt' )
UpperCamelCase_ = self.model_abit.generate(input_ids=encoded_input['input_ids'].to(0 ) , max_new_tokens=10 )
self.assertIn(self.tokenizer.decode(output_sequences[0] , skip_special_tokens=_UpperCAmelCase ) , self.EXPECTED_OUTPUTS )
def _UpperCAmelCase ( self ) -> Tuple:
UpperCamelCase_ = BitsAndBytesConfig()
UpperCamelCase_ = True
UpperCamelCase_ = AutoModelForCausalLM.from_pretrained(
self.model_name , quantization_config=_UpperCAmelCase , device_map='auto' )
UpperCamelCase_ = self.tokenizer(self.input_text , return_tensors='pt' )
UpperCamelCase_ = model_abit_from_config.generate(
input_ids=encoded_input['input_ids'].to(0 ) , max_new_tokens=10 )
self.assertIn(self.tokenizer.decode(output_sequences[0] , skip_special_tokens=_UpperCAmelCase ) , self.EXPECTED_OUTPUTS )
def _UpperCAmelCase ( self ) -> int:
with self.assertRaises(_UpperCAmelCase ), tempfile.TemporaryDirectory() as tmpdirname:
self.model_abit.save_pretrained(_UpperCAmelCase )
def _UpperCAmelCase ( self ) -> List[Any]:
UpperCamelCase_ = BitsAndBytesConfig()
with self.assertRaises(_UpperCAmelCase ):
UpperCamelCase_ = AutoModelForCausalLM.from_pretrained(
self.model_name , quantization_config=_UpperCAmelCase , load_in_abit=_UpperCAmelCase , device_map='auto' , bnb_abit_quant_type='nf4' , )
def _UpperCAmelCase ( self ) -> Optional[Any]:
with self.assertRaises(_UpperCAmelCase ):
# Tries with `str`
self.model_abit.to('cpu' )
with self.assertRaises(_UpperCAmelCase ):
# Tries with a `dtype``
self.model_abit.to(torch.floataa )
with self.assertRaises(_UpperCAmelCase ):
# Tries with a `device`
self.model_abit.to(torch.device('cuda:0' ) )
with self.assertRaises(_UpperCAmelCase ):
# Tries with a `device`
self.model_abit.float()
with self.assertRaises(_UpperCAmelCase ):
# Tries with a `device`
self.model_abit.half()
# Test if we did not break anything
UpperCamelCase_ = self.tokenizer(self.input_text , return_tensors='pt' )
UpperCamelCase_ = self.model_fpaa.to(torch.floataa )
UpperCamelCase_ = self.model_fpaa.generate(input_ids=encoded_input['input_ids'].to(0 ) , max_new_tokens=10 )
# Check this does not throw an error
UpperCamelCase_ = self.model_fpaa.to('cpu' )
# Check this does not throw an error
UpperCamelCase_ = self.model_fpaa.half()
# Check this does not throw an error
UpperCamelCase_ = self.model_fpaa.float()
def _UpperCAmelCase ( self ) -> str:
UpperCamelCase_ = AutoModelForSeqaSeqLM.from_pretrained('t5-small' , load_in_abit=_UpperCAmelCase , device_map='auto' )
self.assertTrue(model.decoder.block[0].layer[2].DenseReluDense.wo.weight.dtype == torch.floataa )
@require_bitsandbytes
@require_accelerate
@require_torch
@require_torch_gpu
@slow
class _a ( unittest.TestCase ):
"""simple docstring"""
@classmethod
def _UpperCAmelCase ( cls ) -> Tuple:
UpperCamelCase_ = 't5-small'
UpperCamelCase_ = 'google/flan-t5-small' # flan-t5 uses dense-act instead of dense-relu-dense
UpperCamelCase_ = AutoTokenizer.from_pretrained(cls.model_name )
UpperCamelCase_ = 'Translate in German: Hello, my dog is cute'
def _UpperCAmelCase ( self ) -> List[Any]:
gc.collect()
torch.cuda.empty_cache()
def _UpperCAmelCase ( self ) -> Any:
from transformers import TaForConditionalGeneration
UpperCamelCase_ = TaForConditionalGeneration._keep_in_fpaa_modules
UpperCamelCase_ = None
# test with `t5-small`
UpperCamelCase_ = TaForConditionalGeneration.from_pretrained(self.model_name , load_in_abit=_UpperCAmelCase , device_map='auto' )
UpperCamelCase_ = self.tokenizer(self.input_text , return_tensors='pt' ).to(0 )
UpperCamelCase_ = model.generate(**_UpperCAmelCase )
# test with `flan-t5-small`
UpperCamelCase_ = TaForConditionalGeneration.from_pretrained(
self.dense_act_model_name , load_in_abit=_UpperCAmelCase , device_map='auto' )
UpperCamelCase_ = self.tokenizer(self.input_text , return_tensors='pt' ).to(0 )
UpperCamelCase_ = model.generate(**_UpperCAmelCase )
UpperCamelCase_ = modules
def _UpperCAmelCase ( self ) -> str:
import bitsandbytes as bnb
from transformers import TaForConditionalGeneration
# test with `t5-small`
UpperCamelCase_ = TaForConditionalGeneration.from_pretrained(self.model_name , load_in_abit=_UpperCAmelCase , device_map='auto' )
# there was a bug with decoders - this test checks that it is fixed
self.assertTrue(isinstance(model.decoder.block[0].layer[0].SelfAttention.q , bnb.nn.Linearabit ) )
UpperCamelCase_ = self.tokenizer(self.input_text , return_tensors='pt' ).to(0 )
UpperCamelCase_ = model.generate(**_UpperCAmelCase )
# test with `flan-t5-small`
UpperCamelCase_ = TaForConditionalGeneration.from_pretrained(
self.dense_act_model_name , load_in_abit=_UpperCAmelCase , device_map='auto' )
UpperCamelCase_ = self.tokenizer(self.input_text , return_tensors='pt' ).to(0 )
UpperCamelCase_ = model.generate(**_UpperCAmelCase )
class _a ( UpperCAmelCase__ ):
"""simple docstring"""
def _UpperCAmelCase ( self ) -> Union[str, Any]:
super().setUp()
# model_name
UpperCamelCase_ = 'bigscience/bloom-560m'
UpperCamelCase_ = 't5-small'
# Different types of model
UpperCamelCase_ = AutoModel.from_pretrained(self.model_name , load_in_abit=_UpperCAmelCase , device_map='auto' )
# Sequence classification model
UpperCamelCase_ = AutoModelForSequenceClassification.from_pretrained(
self.model_name , load_in_abit=_UpperCAmelCase , device_map='auto' )
# CausalLM model
UpperCamelCase_ = AutoModelForCausalLM.from_pretrained(self.model_name , load_in_abit=_UpperCAmelCase , device_map='auto' )
# Seq2seq model
UpperCamelCase_ = AutoModelForSeqaSeqLM.from_pretrained(
self.seq_to_seq_name , load_in_abit=_UpperCAmelCase , device_map='auto' )
def _UpperCAmelCase ( self ) -> Union[str, Any]:
del self.base_model
del self.sequence_model
del self.model_abit
del self.seq_to_seq_model
gc.collect()
torch.cuda.empty_cache()
def _UpperCAmelCase ( self ) -> Union[str, Any]:
from bitsandbytes.nn import Paramsabit
self.assertTrue(self.base_model.h[-1].mlp.dense_ah_to_h.weight.__class__ == Paramsabit )
# Other heads should be nn.Parameter
self.assertTrue(self.model_abit.lm_head.weight.__class__ == torch.nn.Parameter )
self.assertTrue(self.sequence_model.score.weight.__class__ == torch.nn.Parameter )
self.assertTrue(self.seq_to_seq_model.lm_head.weight.__class__ == torch.nn.Parameter )
class _a ( UpperCAmelCase__ ):
"""simple docstring"""
def _UpperCAmelCase ( self ) -> Any:
super().setUp()
def _UpperCAmelCase ( self ) -> Optional[Any]:
del self.pipe
gc.collect()
torch.cuda.empty_cache()
def _UpperCAmelCase ( self ) -> Any:
UpperCamelCase_ = pipeline(
'text-generation' , model=self.model_name , model_kwargs={'device_map': 'auto', 'load_in_4bit': True, 'torch_dtype': torch.floataa} , max_new_tokens=self.MAX_NEW_TOKENS , )
# Real second forward pass
UpperCamelCase_ = self.pipe(self.input_text )
self.assertIn(pipeline_output[0]['generated_text'] , self.EXPECTED_OUTPUTS )
@require_torch_multi_gpu
class _a ( UpperCAmelCase__ ):
"""simple docstring"""
def _UpperCAmelCase ( self ) -> Dict:
super().setUp()
def _UpperCAmelCase ( self ) -> int:
UpperCamelCase_ = AutoModelForCausalLM.from_pretrained(
self.model_name , load_in_abit=_UpperCAmelCase , device_map='balanced' )
# Check correct device map
self.assertEqual(set(model_parallel.hf_device_map.values() ) , {0, 1} )
# Check that inference pass works on the model
UpperCamelCase_ = self.tokenizer(self.input_text , return_tensors='pt' )
# Second real batch
UpperCamelCase_ = model_parallel.generate(input_ids=encoded_input['input_ids'].to(0 ) , max_new_tokens=10 )
self.assertIn(self.tokenizer.decode(output_parallel[0] , skip_special_tokens=_UpperCAmelCase ) , self.EXPECTED_OUTPUTS )
class _a ( UpperCAmelCase__ ):
"""simple docstring"""
def _UpperCAmelCase ( self ) -> str:
UpperCamelCase_ = 'facebook/opt-350m'
super().setUp()
def _UpperCAmelCase ( self ) -> Optional[Any]:
if version.parse(importlib.metadata.version('bitsandbytes' ) ) < version.parse('0.37.0' ):
return
# Step 1: freeze all parameters
UpperCamelCase_ = AutoModelForCausalLM.from_pretrained(self.model_name , load_in_abit=_UpperCAmelCase )
self.assertEqual(set(model.hf_device_map.values() ) , {torch.cuda.current_device()} )
for param in model.parameters():
UpperCamelCase_ = False # freeze the model - train adapters later
if param.ndim == 1:
# cast the small parameters (e.g. layernorm) to fp32 for stability
UpperCamelCase_ = param.data.to(torch.floataa )
# Step 2: add adapters
for _, module in model.named_modules():
if "OPTAttention" in repr(type(_UpperCAmelCase ) ):
UpperCamelCase_ = LoRALayer(module.q_proj , rank=16 )
UpperCamelCase_ = LoRALayer(module.k_proj , rank=16 )
UpperCamelCase_ = LoRALayer(module.v_proj , rank=16 )
# Step 3: dummy batch
UpperCamelCase_ = self.tokenizer('Test batch ' , return_tensors='pt' ).to(0 )
# Step 4: Check if the gradient is not None
with torch.cuda.amp.autocast():
UpperCamelCase_ = model.forward(**_UpperCAmelCase )
out.logits.norm().backward()
for module in model.modules():
if isinstance(_UpperCAmelCase , _UpperCAmelCase ):
self.assertTrue(module.adapter[1].weight.grad is not None )
self.assertTrue(module.adapter[1].weight.grad.norm().item() > 0 )
elif isinstance(_UpperCAmelCase , nn.Embedding ):
self.assertTrue(module.weight.grad is None )
class _a ( UpperCAmelCase__ ):
"""simple docstring"""
A_ = """gpt2-xl"""
A_ = 3.3_191_854_854_152_187
| 23 |
from ..utils import DummyObject, requires_backends
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Optional[Any] = ['torch', 'transformers', 'onnx']
def __init__( self : str , *UpperCAmelCase : int , **UpperCAmelCase : List[Any] ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Dict , *UpperCAmelCase : Dict , **UpperCAmelCase : Union[str, Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[Any] , *UpperCAmelCase : List[Any] , **UpperCAmelCase : Union[str, Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : str = ['torch', 'transformers', 'onnx']
def __init__( self : Optional[int] , *UpperCAmelCase : List[str] , **UpperCAmelCase : int ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : List[str] , *UpperCAmelCase : List[str] , **UpperCAmelCase : List[Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[Any] , *UpperCAmelCase : Dict , **UpperCAmelCase : Optional[Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Optional[Any] = ['torch', 'transformers', 'onnx']
def __init__( self : Union[str, Any] , *UpperCAmelCase : Any , **UpperCAmelCase : Dict ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Tuple , *UpperCAmelCase : Dict , **UpperCAmelCase : str ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Dict , *UpperCAmelCase : List[str] , **UpperCAmelCase : List[str] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : int = ['torch', 'transformers', 'onnx']
def __init__( self : List[str] , *UpperCAmelCase : Dict , **UpperCAmelCase : Optional[Any] ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Any , *UpperCAmelCase : Union[str, Any] , **UpperCAmelCase : Dict ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[Any] , *UpperCAmelCase : int , **UpperCAmelCase : Optional[int] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Dict = ['torch', 'transformers', 'onnx']
def __init__( self : List[str] , *UpperCAmelCase : str , **UpperCAmelCase : int ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : List[str] , *UpperCAmelCase : Optional[int] , **UpperCAmelCase : Optional[int] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[int] , *UpperCAmelCase : int , **UpperCAmelCase : List[Any] ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
class _a ( metaclass=snake_case_ ):
"""simple docstring"""
_lowerCamelCase : int = ['torch', 'transformers', 'onnx']
def __init__( self : Tuple , *UpperCAmelCase : List[Any] , **UpperCAmelCase : Optional[Any] ):
requires_backends(self , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[int] , *UpperCAmelCase : Dict , **UpperCAmelCase : str ):
requires_backends(cls , ["torch", "transformers", "onnx"] )
@classmethod
def __A ( cls : Optional[int] , *UpperCAmelCase : Optional[Any] , **UpperCAmelCase : int ):
requires_backends(cls , ["torch", "transformers", "onnx"] ) | 86 | 0 |
'''simple docstring'''
import json
import os
import tempfile
import datasets
from utils import generate_example_dataset, get_duration
UpperCAmelCase_ : List[Any] = 5_0_0_0_0
UpperCAmelCase_ : Optional[Any] = 5_0_0_0
UpperCAmelCase_ , UpperCAmelCase_ : Optional[int] = os.path.split(__file__)
UpperCAmelCase_ : Tuple = os.path.join(RESULTS_BASEPATH, '''results''', RESULTS_FILENAME.replace('''.py''', '''.json'''))
@get_duration
def _UpperCamelCase (_lowerCamelCase : datasets.Dataset , _lowerCamelCase : Optional[int] )-> Optional[int]:
'''simple docstring'''
for i in range(_lowerCamelCase ):
__snake_case = dataset[i]
@get_duration
def _UpperCamelCase (_lowerCamelCase : datasets.Dataset , _lowerCamelCase : List[str] , _lowerCamelCase : List[str] )-> str:
'''simple docstring'''
for i in range(0 , len(_lowerCamelCase ) , _lowerCamelCase ):
__snake_case = dataset[i : i + batch_size]
@get_duration
def _UpperCamelCase (_lowerCamelCase : datasets.Dataset , _lowerCamelCase : int , _lowerCamelCase : Dict )-> str:
'''simple docstring'''
with dataset.formatted_as(type=_lowerCamelCase ):
for i in range(_lowerCamelCase ):
__snake_case = dataset[i]
@get_duration
def _UpperCamelCase (_lowerCamelCase : datasets.Dataset , _lowerCamelCase : str , _lowerCamelCase : Optional[Any] , _lowerCamelCase : Tuple )-> str:
'''simple docstring'''
with dataset.formatted_as(type=_lowerCamelCase ):
for i in range(0 , _lowerCamelCase , _lowerCamelCase ):
__snake_case = dataset[i : i + batch_size]
def _UpperCamelCase ()-> Tuple:
'''simple docstring'''
__snake_case = {'''num examples''': SPEED_TEST_N_EXAMPLES}
__snake_case = [
(read, {'''length''': SMALL_TEST}),
(read, {'''length''': SPEED_TEST_N_EXAMPLES}),
(read_batch, {'''length''': SPEED_TEST_N_EXAMPLES, '''batch_size''': 10}),
(read_batch, {'''length''': SPEED_TEST_N_EXAMPLES, '''batch_size''': 1_00}),
(read_batch, {'''length''': SPEED_TEST_N_EXAMPLES, '''batch_size''': 10_00}),
(read_formatted, {'''type''': '''numpy''', '''length''': SMALL_TEST}),
(read_formatted, {'''type''': '''pandas''', '''length''': SMALL_TEST}),
(read_formatted, {'''type''': '''torch''', '''length''': SMALL_TEST}),
(read_formatted, {'''type''': '''tensorflow''', '''length''': SMALL_TEST}),
(read_formatted_batch, {'''type''': '''numpy''', '''length''': SMALL_TEST, '''batch_size''': 10}),
(read_formatted_batch, {'''type''': '''numpy''', '''length''': SMALL_TEST, '''batch_size''': 10_00}),
]
__snake_case = [
(read, {'''length''': SMALL_TEST}),
(read, {'''length''': SPEED_TEST_N_EXAMPLES}),
(read_batch, {'''length''': SPEED_TEST_N_EXAMPLES, '''batch_size''': 10}),
(read_batch, {'''length''': SPEED_TEST_N_EXAMPLES, '''batch_size''': 1_00}),
(read_batch, {'''length''': SPEED_TEST_N_EXAMPLES, '''batch_size''': 10_00}),
(read_formatted, {'''type''': '''numpy''', '''length''': SMALL_TEST}),
(read_formatted_batch, {'''type''': '''numpy''', '''length''': SMALL_TEST, '''batch_size''': 10}),
(read_formatted_batch, {'''type''': '''numpy''', '''length''': SMALL_TEST, '''batch_size''': 10_00}),
]
with tempfile.TemporaryDirectory() as tmp_dir:
print('''generating dataset''' )
__snake_case = datasets.Features(
{'''list''': datasets.Sequence(datasets.Value('''float32''' ) ), '''numbers''': datasets.Value('''float32''' )} )
__snake_case = generate_example_dataset(
os.path.join(_lowerCamelCase , '''dataset.arrow''' ) , _lowerCamelCase , num_examples=_lowerCamelCase , seq_shapes={'''list''': (1_00,)} , )
print('''first set of iterations''' )
for func, kwargs in functions:
print(func.__name__ , str(_lowerCamelCase ) )
__snake_case = func(_lowerCamelCase , **_lowerCamelCase )
print('''shuffling dataset''' )
__snake_case = dataset.shuffle()
print('''Second set of iterations (after shuffling''' )
for func, kwargs in functions_shuffled:
print('''shuffled ''' , func.__name__ , str(_lowerCamelCase ) )
__snake_case = func(
_lowerCamelCase , **_lowerCamelCase )
with open(_lowerCamelCase , '''wb''' ) as f:
f.write(json.dumps(_lowerCamelCase ).encode('''utf-8''' ) )
if __name__ == "__main__": # useful to run the profiler
benchmark_iterating()
| 24 |
import torch
from diffusers import DDPMParallelScheduler
from .test_schedulers import SchedulerCommonTest
class _a ( snake_case_ ):
"""simple docstring"""
_lowerCamelCase : Optional[Any] = (DDPMParallelScheduler,)
def __A ( self : List[Any] , **UpperCAmelCase : Optional[int] ):
A_ = {
"num_train_timesteps": 1000,
"beta_start": 0.0_001,
"beta_end": 0.02,
"beta_schedule": "linear",
"variance_type": "fixed_small",
"clip_sample": True,
}
config.update(**UpperCAmelCase )
return config
def __A ( self : Optional[Any] ):
for timesteps in [1, 5, 100, 1000]:
self.check_over_configs(num_train_timesteps=UpperCAmelCase )
def __A ( self : Dict ):
for beta_start, beta_end in zip([0.0_001, 0.001, 0.01, 0.1] , [0.002, 0.02, 0.2, 2] ):
self.check_over_configs(beta_start=UpperCAmelCase , beta_end=UpperCAmelCase )
def __A ( self : int ):
for schedule in ["linear", "squaredcos_cap_v2"]:
self.check_over_configs(beta_schedule=UpperCAmelCase )
def __A ( self : Tuple ):
for variance in ["fixed_small", "fixed_large", "other"]:
self.check_over_configs(variance_type=UpperCAmelCase )
def __A ( self : int ):
for clip_sample in [True, False]:
self.check_over_configs(clip_sample=UpperCAmelCase )
def __A ( self : Union[str, Any] ):
self.check_over_configs(thresholding=UpperCAmelCase )
for threshold in [0.5, 1.0, 2.0]:
for prediction_type in ["epsilon", "sample", "v_prediction"]:
self.check_over_configs(
thresholding=UpperCAmelCase , prediction_type=UpperCAmelCase , sample_max_value=UpperCAmelCase , )
def __A ( self : Optional[int] ):
for prediction_type in ["epsilon", "sample", "v_prediction"]:
self.check_over_configs(prediction_type=UpperCAmelCase )
def __A ( self : Tuple ):
for t in [0, 500, 999]:
self.check_over_forward(time_step=UpperCAmelCase )
def __A ( self : Tuple ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
assert torch.sum(torch.abs(scheduler._get_variance(0 ) - 0.0 ) ) < 1E-5
assert torch.sum(torch.abs(scheduler._get_variance(487 ) - 0.00_979 ) ) < 1E-5
assert torch.sum(torch.abs(scheduler._get_variance(999 ) - 0.02 ) ) < 1E-5
def __A ( self : List[Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = len(UpperCAmelCase )
A_ = self.dummy_model()
A_ = self.dummy_sample_deter
A_ = self.dummy_sample_deter + 0.1
A_ = self.dummy_sample_deter - 0.1
A_ = samplea.shape[0]
A_ = torch.stack([samplea, samplea, samplea] , dim=0 )
A_ = torch.arange(UpperCAmelCase )[0:3, None].repeat(1 , UpperCAmelCase )
A_ = model(samples.flatten(0 , 1 ) , timesteps.flatten(0 , 1 ) )
A_ = scheduler.batch_step_no_noise(UpperCAmelCase , timesteps.flatten(0 , 1 ) , samples.flatten(0 , 1 ) )
A_ = torch.sum(torch.abs(UpperCAmelCase ) )
A_ = torch.mean(torch.abs(UpperCAmelCase ) )
assert abs(result_sum.item() - 1_153.1_833 ) < 1E-2
assert abs(result_mean.item() - 0.5_005 ) < 1E-3
def __A ( self : Tuple ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = len(UpperCAmelCase )
A_ = self.dummy_model()
A_ = self.dummy_sample_deter
A_ = torch.manual_seed(0 )
for t in reversed(range(UpperCAmelCase ) ):
# 1. predict noise residual
A_ = model(UpperCAmelCase , UpperCAmelCase )
# 2. predict previous mean of sample x_t-1
A_ = scheduler.step(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase , generator=UpperCAmelCase ).prev_sample
A_ = pred_prev_sample
A_ = torch.sum(torch.abs(UpperCAmelCase ) )
A_ = torch.mean(torch.abs(UpperCAmelCase ) )
assert abs(result_sum.item() - 258.9_606 ) < 1E-2
assert abs(result_mean.item() - 0.3_372 ) < 1E-3
def __A ( self : Tuple ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config(prediction_type="v_prediction" )
A_ = scheduler_class(**UpperCAmelCase )
A_ = len(UpperCAmelCase )
A_ = self.dummy_model()
A_ = self.dummy_sample_deter
A_ = torch.manual_seed(0 )
for t in reversed(range(UpperCAmelCase ) ):
# 1. predict noise residual
A_ = model(UpperCAmelCase , UpperCAmelCase )
# 2. predict previous mean of sample x_t-1
A_ = scheduler.step(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase , generator=UpperCAmelCase ).prev_sample
A_ = pred_prev_sample
A_ = torch.sum(torch.abs(UpperCAmelCase ) )
A_ = torch.mean(torch.abs(UpperCAmelCase ) )
assert abs(result_sum.item() - 202.0_296 ) < 1E-2
assert abs(result_mean.item() - 0.2_631 ) < 1E-3
def __A ( self : Union[str, Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = [100, 87, 50, 1, 0]
scheduler.set_timesteps(timesteps=UpperCAmelCase )
A_ = scheduler.timesteps
for i, timestep in enumerate(UpperCAmelCase ):
if i == len(UpperCAmelCase ) - 1:
A_ = -1
else:
A_ = timesteps[i + 1]
A_ = scheduler.previous_timestep(UpperCAmelCase )
A_ = prev_t.item()
self.assertEqual(UpperCAmelCase , UpperCAmelCase )
def __A ( self : List[Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = [100, 87, 50, 51, 0]
with self.assertRaises(UpperCAmelCase , msg="`custom_timesteps` must be in descending order." ):
scheduler.set_timesteps(timesteps=UpperCAmelCase )
def __A ( self : List[Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = [100, 87, 50, 1, 0]
A_ = len(UpperCAmelCase )
with self.assertRaises(UpperCAmelCase , msg="Can only pass one of `num_inference_steps` or `custom_timesteps`." ):
scheduler.set_timesteps(num_inference_steps=UpperCAmelCase , timesteps=UpperCAmelCase )
def __A ( self : Optional[Any] ):
A_ = self.scheduler_classes[0]
A_ = self.get_scheduler_config()
A_ = scheduler_class(**UpperCAmelCase )
A_ = [scheduler.config.num_train_timesteps]
with self.assertRaises(
UpperCAmelCase , msg="`timesteps` must start before `self.config.train_timesteps`: {scheduler.config.num_train_timesteps}}" , ):
scheduler.set_timesteps(timesteps=UpperCAmelCase ) | 86 | 0 |
import tempfile
import unittest
from transformers import TaConfig, is_torch_available
from transformers.testing_utils import (
require_sentencepiece,
require_tokenizers,
require_torch,
slow,
torch_device,
)
from ...generation.test_utils import GenerationTesterMixin
from ...test_modeling_common import ModelTesterMixin, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from transformers import AutoTokenizer, UMTaForConditionalGeneration, UMTaForQuestionAnswering, UMTaModel
class _UpperCamelCase :
'''simple docstring'''
def __init__( self : List[str] , a : Optional[int] , a : str=99 , a : str=13 , a : List[Any]=7 , a : Optional[int]=9 , a : Optional[int]=True , a : Union[str, Any]=True , a : Any=False , a : Tuple=32 , a : List[str]=5 , a : Union[str, Any]=4 , a : Union[str, Any]=37 , a : str=8 , a : int=0.1 , a : Optional[int]=0.002 , a : Union[str, Any]=1 , a : Optional[int]=0 , a : Tuple=0 , a : Any=None , a : Optional[Any]=None , ) -> Tuple:
"""simple docstring"""
SCREAMING_SNAKE_CASE : Optional[Any] = parent
SCREAMING_SNAKE_CASE : int = batch_size
SCREAMING_SNAKE_CASE : Optional[int] = encoder_seq_length
SCREAMING_SNAKE_CASE : Tuple = decoder_seq_length
# For common tests
SCREAMING_SNAKE_CASE : int = self.decoder_seq_length
SCREAMING_SNAKE_CASE : int = is_training
SCREAMING_SNAKE_CASE : Any = use_attention_mask
SCREAMING_SNAKE_CASE : List[str] = use_labels
SCREAMING_SNAKE_CASE : Optional[Any] = vocab_size
SCREAMING_SNAKE_CASE : Union[str, Any] = hidden_size
SCREAMING_SNAKE_CASE : Tuple = num_hidden_layers
SCREAMING_SNAKE_CASE : Any = num_attention_heads
SCREAMING_SNAKE_CASE : Optional[int] = d_ff
SCREAMING_SNAKE_CASE : Tuple = relative_attention_num_buckets
SCREAMING_SNAKE_CASE : List[Any] = dropout_rate
SCREAMING_SNAKE_CASE : int = initializer_factor
SCREAMING_SNAKE_CASE : List[Any] = eos_token_id
SCREAMING_SNAKE_CASE : int = pad_token_id
SCREAMING_SNAKE_CASE : Union[str, Any] = decoder_start_token_id
SCREAMING_SNAKE_CASE : Union[str, Any] = None
SCREAMING_SNAKE_CASE : Tuple = decoder_layers
def __UpperCamelCase ( self : Tuple ) -> Optional[Any]:
"""simple docstring"""
return TaConfig.from_pretrained("google/umt5-base" )
def __UpperCamelCase ( self : Optional[int] , a : List[str] , a : Optional[Any] , a : Optional[int] , a : Tuple=None , a : List[Any]=None , a : int=None , a : Any=None , a : Dict=None , ) -> List[str]:
"""simple docstring"""
if attention_mask is None:
SCREAMING_SNAKE_CASE : int = input_ids.ne(config.pad_token_id )
if decoder_attention_mask is None:
SCREAMING_SNAKE_CASE : Any = decoder_input_ids.ne(config.pad_token_id )
if head_mask is None:
SCREAMING_SNAKE_CASE : Tuple = torch.ones(config.num_hidden_layers , config.num_attention_heads , device=a )
if decoder_head_mask is None:
SCREAMING_SNAKE_CASE : Union[str, Any] = torch.ones(config.num_decoder_layers , config.num_attention_heads , device=a )
if cross_attn_head_mask is None:
SCREAMING_SNAKE_CASE : Optional[Any] = torch.ones(
config.num_decoder_layers , config.num_attention_heads , device=a )
return {
"input_ids": input_ids,
"decoder_input_ids": decoder_input_ids,
"attention_mask": attention_mask,
"decoder_attention_mask": decoder_attention_mask,
"head_mask": head_mask,
"decoder_head_mask": decoder_head_mask,
"cross_attn_head_mask": cross_attn_head_mask,
}
def __UpperCamelCase ( self : List[Any] ) -> Union[str, Any]:
"""simple docstring"""
SCREAMING_SNAKE_CASE : int = ids_tensor([self.batch_size, self.encoder_seq_length] , self.vocab_size )
SCREAMING_SNAKE_CASE : Any = ids_tensor([self.batch_size, self.decoder_seq_length] , self.vocab_size )
# we need to clamp the input ids here to avoid having pad token in between
# this is because for NllbMoe the position_ids are prepared such that
# all pad tokens have pos id = 2 and rest are between 2..seq_length
# and the seq_length here is seq_length - num_pad_tokens
# but when using past, there is no way of knowing if the past input ids had
# pad tokens in them, which results in incorrect seq_lenth and which in turn results in
# position_ids being off by num_pad_tokens in past input
SCREAMING_SNAKE_CASE : int = input_ids.clamp(self.pad_token_id + 1 )
SCREAMING_SNAKE_CASE : Tuple = decoder_input_ids.clamp(self.pad_token_id + 1 )
SCREAMING_SNAKE_CASE : Any = self.get_config()
SCREAMING_SNAKE_CASE : Union[str, Any] = config.num_attention_heads
SCREAMING_SNAKE_CASE : Dict = self.prepare_inputs_dict(a , a , a )
return config, input_dict
def __UpperCamelCase ( self : Union[str, Any] ) -> int:
"""simple docstring"""
SCREAMING_SNAKE_CASE ,SCREAMING_SNAKE_CASE : Tuple = self.prepare_config_and_inputs()
return config, inputs_dict
def __UpperCamelCase ( self : int ) -> List[Any]:
"""simple docstring"""
return TaConfig(
vocab_size=166 , d_model=self.hidden_size , d_ff=self.d_ff , d_kv=self.hidden_size // self.num_attention_heads , num_layers=self.num_hidden_layers , num_decoder_layers=self.decoder_layers , num_heads=self.num_attention_heads , relative_attention_num_buckets=self.relative_attention_num_buckets , dropout_rate=self.dropout_rate , initializer_factor=self.initializer_factor , eos_token_id=self.eos_token_id , bos_token_id=self.pad_token_id , pad_token_id=self.pad_token_id , decoder_start_token_id=self.decoder_start_token_id , )
def __UpperCamelCase ( self : List[Any] ) -> int:
"""simple docstring"""
return TaConfig(
vocab_size=self.vocab_size , d_model=self.hidden_size , d_ff=self.d_ff , d_kv=self.hidden_size // self.num_attention_heads , num_layers=self.num_hidden_layers , num_decoder_layers=self.decoder_layers , num_heads=self.num_attention_heads , relative_attention_num_buckets=self.relative_attention_num_buckets , dropout_rate=self.dropout_rate , initializer_factor=self.initializer_factor , eos_token_id=self.eos_token_id , bos_token_id=self.pad_token_id , pad_token_id=self.pad_token_id , decoder_start_token_id=self.decoder_start_token_id , )
def __UpperCamelCase ( self : Dict , a : List[str] , a : Dict , a : int , a : Tuple , a : List[str] , a : Union[str, Any] , ) -> Union[str, Any]:
"""simple docstring"""
SCREAMING_SNAKE_CASE : int = UMTaModel(config=a )
model.to(a )
model.eval()
SCREAMING_SNAKE_CASE : List[str] = model(
input_ids=a , decoder_input_ids=a , attention_mask=a , decoder_attention_mask=a , )
SCREAMING_SNAKE_CASE : Any = model(input_ids=a , decoder_input_ids=a )
SCREAMING_SNAKE_CASE : Optional[Any] = result.last_hidden_state
SCREAMING_SNAKE_CASE : List[str] = result.past_key_values
SCREAMING_SNAKE_CASE : List[str] = result.encoder_last_hidden_state
self.parent.assertEqual(encoder_output.size() , (self.batch_size, self.encoder_seq_length, self.hidden_size) )
self.parent.assertEqual(decoder_output.size() , (self.batch_size, self.decoder_seq_length, self.hidden_size) )
# There should be `num_layers` key value embeddings stored in decoder_past
self.parent.assertEqual(len(a ) , config.num_layers )
# There should be a self attn key, a self attn value, a cross attn key and a cross attn value stored in each decoder_past tuple
self.parent.assertEqual(len(decoder_past[0] ) , 4 )
def __UpperCamelCase ( self : List[str] , a : Optional[int] , a : Tuple , a : Tuple , a : List[Any] , a : List[Any] , a : Optional[int] , ) -> Union[str, Any]:
"""simple docstring"""
SCREAMING_SNAKE_CASE : int = UMTaModel(config=a ).get_decoder().to(a ).eval()
# first forward pass
SCREAMING_SNAKE_CASE : int = model(a , use_cache=a )
SCREAMING_SNAKE_CASE : Dict = model(a )
SCREAMING_SNAKE_CASE : str = model(a , use_cache=a )
self.parent.assertTrue(len(a ) == len(a ) )
self.parent.assertTrue(len(a ) == len(a ) + 1 )
SCREAMING_SNAKE_CASE ,SCREAMING_SNAKE_CASE : Dict = outputs.to_tuple()
# create hypothetical next token and extent to next_input_ids
SCREAMING_SNAKE_CASE : Optional[Any] = ids_tensor((self.batch_size, 1) , config.vocab_size )
# append to next input_ids and
SCREAMING_SNAKE_CASE : List[str] = torch.cat([input_ids, next_tokens] , dim=-1 )
SCREAMING_SNAKE_CASE : str = model(a )["last_hidden_state"]
SCREAMING_SNAKE_CASE : List[str] = model(a , past_key_values=a )["last_hidden_state"]
# select random slice
SCREAMING_SNAKE_CASE : Tuple = ids_tensor((1,) , output_from_past.shape[-1] ).item()
SCREAMING_SNAKE_CASE : List[Any] = output_from_no_past[:, -1, random_slice_idx].detach()
SCREAMING_SNAKE_CASE : Any = output_from_past[:, 0, random_slice_idx].detach()
# test that outputs are equal for slice
self.parent.assertTrue(torch.allclose(a , a , atol=1e-3 ) )
def __UpperCamelCase ( self : Tuple , a : Union[str, Any] , a : int , ) -> int:
"""simple docstring"""
SCREAMING_SNAKE_CASE : str = UMTaModel(config=a ).to(a ).half().eval()
SCREAMING_SNAKE_CASE : Union[str, Any] = model(**a )["last_hidden_state"]
self.parent.assertFalse(torch.isnan(a ).any().item() )
@require_torch
class _UpperCamelCase ( __A , __A , __A , unittest.TestCase ):
'''simple docstring'''
lowerCamelCase__ =(
(UMTaModel, UMTaForConditionalGeneration, UMTaForQuestionAnswering) if is_torch_available() else ()
)
lowerCamelCase__ =(UMTaForConditionalGeneration,) if is_torch_available() else ()
lowerCamelCase__ =(
{
'conversational': UMTaForConditionalGeneration,
'feature-extraction': UMTaModel,
'summarization': UMTaForConditionalGeneration,
'text2text-generation': UMTaForConditionalGeneration,
'translation': UMTaForConditionalGeneration,
'question-answering': UMTaForQuestionAnswering,
}
if is_torch_available()
else {}
)
lowerCamelCase__ =True
lowerCamelCase__ =False
lowerCamelCase__ =False
lowerCamelCase__ =True
lowerCamelCase__ =True
# The small UMT5 model needs higher percentages for CPU/MP tests
lowerCamelCase__ =[0.8, 0.9]
def __UpperCamelCase ( self : Tuple ) -> Tuple:
"""simple docstring"""
SCREAMING_SNAKE_CASE : Dict = UMTaModelTester(self )
@unittest.skip("Test has a segmentation fault on torch 1.8.0" )
def __UpperCamelCase ( self : int ) -> Optional[int]:
"""simple docstring"""
SCREAMING_SNAKE_CASE : str = self.model_tester.prepare_config_and_inputs()
SCREAMING_SNAKE_CASE : int = UMTaModel(config_and_inputs[0] ).to(a )
with tempfile.TemporaryDirectory() as tmpdirname:
torch.onnx.export(
a , (config_and_inputs[1], config_and_inputs[3], config_and_inputs[2]) , F"{tmpdirname}/t5_test.onnx" , export_params=a , opset_version=9 , input_names=["input_ids", "decoder_input_ids"] , )
@unittest.skipIf(torch_device == "cpu" , "Cant do half precision" )
def __UpperCamelCase ( self : Dict ) -> int:
"""simple docstring"""
SCREAMING_SNAKE_CASE : int = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model_fpaa_forward(*a )
def __UpperCamelCase ( self : Tuple ) -> str:
"""simple docstring"""
SCREAMING_SNAKE_CASE : int = ["encoder_attentions", "decoder_attentions", "cross_attentions"]
SCREAMING_SNAKE_CASE : int = self.model_tester.prepare_config_and_inputs()
SCREAMING_SNAKE_CASE : Union[str, Any] = config_and_inputs[0]
SCREAMING_SNAKE_CASE : Any = UMTaForConditionalGeneration(a ).eval()
model.to(a )
SCREAMING_SNAKE_CASE : List[Any] = {
"head_mask": torch.zeros(config.num_layers , config.num_heads , device=a ),
"decoder_head_mask": torch.zeros(config.num_decoder_layers , config.num_heads , device=a ),
"cross_attn_head_mask": torch.zeros(config.num_decoder_layers , config.num_heads , device=a ),
}
for attn_name, (name, mask) in zip(a , head_masking.items() ):
SCREAMING_SNAKE_CASE : Optional[int] = {name: mask}
# Explicitly pass decoder_head_mask as it is required from T5 model when head_mask specified
if name == "head_mask":
SCREAMING_SNAKE_CASE : List[Any] = torch.ones(
config.num_decoder_layers , config.num_heads , device=a )
SCREAMING_SNAKE_CASE : Optional[Any] = model.generate(
config_and_inputs[1]["input_ids"] , num_beams=1 , max_length=3 , output_attentions=a , return_dict_in_generate=a , **a , )
# We check the state of decoder_attentions and cross_attentions just from the last step
SCREAMING_SNAKE_CASE : int = out[attn_name] if attn_name == attention_names[0] else out[attn_name][-1]
self.assertEqual(sum([w.sum().item() for w in attn_weights] ) , 0.0 )
@unittest.skip("Does not work on the tiny model as we keep hitting edge cases." )
def __UpperCamelCase ( self : Optional[int] ) -> int:
"""simple docstring"""
pass
@require_torch
@require_sentencepiece
@require_tokenizers
class _UpperCamelCase ( unittest.TestCase ):
'''simple docstring'''
@slow
@unittest.skip(
"Unless we stop stripping left and right by default for all special tokens, the expected ids obtained here will not match the original ones. Wait for https://github.com/huggingface/transformers/pull/23909 to be merged" )
def __UpperCamelCase ( self : List[str] ) -> str:
"""simple docstring"""
SCREAMING_SNAKE_CASE : str = UMTaForConditionalGeneration.from_pretrained("google/umt5-small" , return_dict=a ).to(a )
SCREAMING_SNAKE_CASE : int = AutoTokenizer.from_pretrained("google/umt5-small" , use_fast=a , legacy=a )
SCREAMING_SNAKE_CASE : int = [
"Bonjour monsieur <extra_id_0> bien <extra_id_1>.",
"No se como puedo <extra_id_0>.",
"This is the reason why we <extra_id_0> them.",
"The <extra_id_0> walks in <extra_id_1>, seats",
"A <extra_id_0> walks into a bar and orders a <extra_id_1> with <extra_id_2> pinch of <extra_id_3>.",
]
SCREAMING_SNAKE_CASE : Dict = tokenizer(a , return_tensors="pt" , padding=a ).input_ids
# fmt: off
SCREAMING_SNAKE_CASE : int = torch.tensor(
[
[ 3_8530, 21_0703, 25_6299, 1410, 25_6298, 274, 1, 0,0, 0, 0, 0, 0, 0, 0, 0,0, 0],
[ 826, 321, 671, 2_5922, 25_6299, 274, 1, 0,0, 0, 0, 0, 0, 0, 0, 0,0, 0],
[ 1460, 339, 312, 1_9014, 1_0620, 758, 25_6299, 2355,274, 1, 0, 0, 0, 0, 0, 0,0, 0],
[ 517, 25_6299, 1_4869, 281, 301, 25_6298, 275, 11_9983,1, 0, 0, 0, 0, 0, 0, 0,0, 0],
[ 320, 25_6299, 1_4869, 281, 2234, 289, 2275, 333,6_1391, 289, 25_6298, 543, 25_6297, 16_8714, 329, 25_6296,274, 1],
] )
# fmt: on
torch.testing.assert_allclose(a , a )
SCREAMING_SNAKE_CASE : Dict = model.generate(input_ids.to(a ) )
SCREAMING_SNAKE_CASE : List[Any] = [
"<pad><extra_id_0> et<extra_id_1> [eod] <extra_id_2><extra_id_55>.. [eod] 💐 💐 💐 💐 💐 💐 💐 💐 💐 💐 💐 <extra_id_56>ajšietosto<extra_id_56>lleux<extra_id_19><extra_id_6>ajšie</s>",
"<pad><extra_id_0>.<extra_id_1>.,<0x0A>...spech <0x0A><extra_id_20> <extra_id_21></s><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad>",
"<pad><extra_id_0> are not going to be a part of the world. We are not going to be a part of<extra_id_1> and<extra_id_2><0x0A><extra_id_48>.<extra_id_48></s><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad>",
"<pad><extra_id_0> door<extra_id_1>, the door<extra_id_2> 피해[/</s><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad>",
"<pad><extra_id_0>nyone who<extra_id_1> drink<extra_id_2> a<extra_id_3> alcohol<extra_id_4> A<extra_id_5> A. This<extra_id_6> I<extra_id_7><extra_id_52><extra_id_53></s><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad>",
]
SCREAMING_SNAKE_CASE : Dict = tokenizer.batch_decode(a )
self.assertEqual(a , a ) | 25 |
import argparse
import json
import os
import torch
from transformers import LukeConfig, LukeModel, LukeTokenizer, RobertaTokenizer
from transformers.tokenization_utils_base import AddedToken
@torch.no_grad()
def __snake_case ( __UpperCamelCase : Tuple ,__UpperCamelCase : Dict ,__UpperCamelCase : Union[str, Any] ,__UpperCamelCase : List[Any] ,__UpperCamelCase : List[Any] ):
"""simple docstring"""
with open(__UpperCamelCase ) as metadata_file:
A_ = json.load(__UpperCamelCase )
A_ = LukeConfig(use_entity_aware_attention=__UpperCamelCase ,**metadata["model_config"] )
# Load in the weights from the checkpoint_path
A_ = torch.load(__UpperCamelCase ,map_location="cpu" )
# Load the entity vocab file
A_ = load_entity_vocab(__UpperCamelCase )
A_ = RobertaTokenizer.from_pretrained(metadata["model_config"]["bert_model_name"] )
# Add special tokens to the token vocabulary for downstream tasks
A_ = AddedToken("<ent>" ,lstrip=__UpperCamelCase ,rstrip=__UpperCamelCase )
A_ = AddedToken("<ent2>" ,lstrip=__UpperCamelCase ,rstrip=__UpperCamelCase )
tokenizer.add_special_tokens({"additional_special_tokens": [entity_token_a, entity_token_a]} )
config.vocab_size += 2
print(f'''Saving tokenizer to {pytorch_dump_folder_path}''' )
tokenizer.save_pretrained(__UpperCamelCase )
with open(os.path.join(__UpperCamelCase ,LukeTokenizer.vocab_files_names["entity_vocab_file"] ) ,"w" ) as f:
json.dump(__UpperCamelCase ,__UpperCamelCase )
A_ = LukeTokenizer.from_pretrained(__UpperCamelCase )
# Initialize the embeddings of the special tokens
A_ = state_dict["embeddings.word_embeddings.weight"]
A_ = word_emb[tokenizer.convert_tokens_to_ids(["@"] )[0]].unsqueeze(0 )
A_ = word_emb[tokenizer.convert_tokens_to_ids(["#"] )[0]].unsqueeze(0 )
A_ = torch.cat([word_emb, ent_emb, enta_emb] )
# Initialize the query layers of the entity-aware self-attention mechanism
for layer_index in range(config.num_hidden_layers ):
for matrix_name in ["query.weight", "query.bias"]:
A_ = f'''encoder.layer.{layer_index}.attention.self.'''
A_ = state_dict[prefix + matrix_name]
A_ = state_dict[prefix + matrix_name]
A_ = state_dict[prefix + matrix_name]
# Initialize the embedding of the [MASK2] entity using that of the [MASK] entity for downstream tasks
A_ = state_dict["entity_embeddings.entity_embeddings.weight"]
A_ = entity_emb[entity_vocab["[MASK]"]]
A_ = LukeModel(config=__UpperCamelCase ).eval()
A_ , A_ = model.load_state_dict(__UpperCamelCase ,strict=__UpperCamelCase )
if not (len(__UpperCamelCase ) == 1 and missing_keys[0] == "embeddings.position_ids"):
raise ValueError(f'''Missing keys {", ".join(__UpperCamelCase )}. Expected only missing embeddings.position_ids''' )
if not (all(key.startswith("entity_predictions" ) or key.startswith("lm_head" ) for key in unexpected_keys )):
raise ValueError(
"Unexpected keys"
f''' {", ".join([key for key in unexpected_keys if not (key.startswith("entity_predictions" ) or key.startswith("lm_head" ))] )}''' )
# Check outputs
A_ = LukeTokenizer.from_pretrained(__UpperCamelCase ,task="entity_classification" )
A_ = (
"Top seed Ana Ivanovic said on Thursday she could hardly believe her luck as a fortuitous netcord helped the"
" new world number one avoid a humiliating second- round exit at Wimbledon ."
)
A_ = (39, 42)
A_ = tokenizer(__UpperCamelCase ,entity_spans=[span] ,add_prefix_space=__UpperCamelCase ,return_tensors="pt" )
A_ = model(**__UpperCamelCase )
# Verify word hidden states
if model_size == "large":
A_ = torch.Size((1, 42, 1024) )
A_ = torch.tensor(
[[0.0133, 0.0865, 0.0095], [0.3093, -0.2576, -0.7418], [-0.1720, -0.2117, -0.2869]] )
else: # base
A_ = torch.Size((1, 42, 768) )
A_ = torch.tensor([[0.0037, 0.1368, -0.0091], [0.1099, 0.3329, -0.1095], [0.0765, 0.5335, 0.1179]] )
if not (outputs.last_hidden_state.shape == expected_shape):
raise ValueError(
f'''Outputs.last_hidden_state.shape is {outputs.last_hidden_state.shape}, Expected shape is {expected_shape}''' )
if not torch.allclose(outputs.last_hidden_state[0, :3, :3] ,__UpperCamelCase ,atol=1E-4 ):
raise ValueError
# Verify entity hidden states
if model_size == "large":
A_ = torch.Size((1, 1, 1024) )
A_ = torch.tensor([[0.0466, -0.0106, -0.0179]] )
else: # base
A_ = torch.Size((1, 1, 768) )
A_ = torch.tensor([[0.1457, 0.1044, 0.0174]] )
if not (outputs.entity_last_hidden_state.shape != expected_shape):
raise ValueError(
f'''Outputs.entity_last_hidden_state.shape is {outputs.entity_last_hidden_state.shape}, Expected shape is'''
f''' {expected_shape}''' )
if not torch.allclose(outputs.entity_last_hidden_state[0, :3, :3] ,__UpperCamelCase ,atol=1E-4 ):
raise ValueError
# Finally, save our PyTorch model and tokenizer
print("Saving PyTorch model to {}".format(__UpperCamelCase ) )
model.save_pretrained(__UpperCamelCase )
def __snake_case ( __UpperCamelCase : str ):
"""simple docstring"""
A_ = {}
with open(__UpperCamelCase ,"r" ,encoding="utf-8" ) as f:
for index, line in enumerate(__UpperCamelCase ):
A_ , A_ = line.rstrip().split("\t" )
A_ = index
return entity_vocab
if __name__ == "__main__":
__a :Dict = argparse.ArgumentParser()
# Required parameters
parser.add_argument('--checkpoint_path', type=str, help='Path to a pytorch_model.bin file.')
parser.add_argument(
'--metadata_path', default=None, type=str, help='Path to a metadata.json file, defining the configuration.'
)
parser.add_argument(
'--entity_vocab_path',
default=None,
type=str,
help='Path to an entity_vocab.tsv file, containing the entity vocabulary.',
)
parser.add_argument(
'--pytorch_dump_folder_path', default=None, type=str, help='Path to where to dump the output PyTorch model.'
)
parser.add_argument(
'--model_size', default='base', type=str, choices=['base', 'large'], help='Size of the model to be converted.'
)
__a :Tuple = parser.parse_args()
convert_luke_checkpoint(
args.checkpoint_path,
args.metadata_path,
args.entity_vocab_path,
args.pytorch_dump_folder_path,
args.model_size,
) | 86 | 0 |
'''simple docstring'''
import logging
import sys
from dataclasses import dataclass, field
from typing import Any, Dict, List, Optional, Union
import librosa
import torch
from datasets import DatasetDict, load_dataset
from packaging import version
from torch import nn
from transformers import (
HfArgumentParser,
Trainer,
TrainingArguments,
WavaVecaConfig,
WavaVecaFeatureExtractor,
WavaVecaForPreTraining,
is_apex_available,
trainer_utils,
)
from transformers.models.wavaveca.modeling_wavaveca import _compute_mask_indices
if is_apex_available():
from apex import amp
if version.parse(version.parse(torch.__version__).base_version) >= version.parse("1.6"):
__UpperCamelCase = True
from torch.cuda.amp import autocast
__UpperCamelCase = logging.getLogger(__name__)
@dataclass
class _A :
lowercase__: str = field(
metadata={'''help''': '''Path to pretrained model or model identifier from huggingface.co/models'''} )
lowercase__: Optional[str] = field(
default=__lowercase , metadata={'''help''': '''Where do you want to store the pretrained models downloaded from huggingface.co'''} , )
lowercase__: Optional[bool] = field(
default=__lowercase , metadata={'''help''': '''Whether to freeze the feature extractor layers of the model.'''} )
lowercase__: Optional[bool] = field(
default=__lowercase , metadata={'''help''': '''Whether to log verbose messages or not.'''} , )
lowercase__: Optional[float] = field(
default=2.0 , metadata={'''help''': '''Maximum temperature for gumbel softmax.'''} )
lowercase__: Optional[float] = field(
default=0.5 , metadata={'''help''': '''Minimum temperature for gumbel softmax.'''} )
lowercase__: Optional[float] = field(
default=0.9_9_9_9_9_5 , metadata={'''help''': '''Decay of gumbel temperature during training.'''} )
def _a ( _lowerCamelCase , _lowerCamelCase ) -> Optional[int]:
"""simple docstring"""
logging.basicConfig(
format="""%(asctime)s - %(levelname)s - %(name)s - %(message)s""" , datefmt="""%m/%d/%Y %H:%M:%S""" , handlers=[logging.StreamHandler(sys.stdout )] , )
__snake_case : int = logging.WARNING
if model_args.verbose_logging:
__snake_case : Any = logging.DEBUG
elif trainer_utils.is_main_process(training_args.local_rank ):
__snake_case : List[str] = logging.INFO
logger.setLevel(_lowerCamelCase )
@dataclass
class _A :
lowercase__: str = field(
default=__lowercase , metadata={'''help''': '''The name of the dataset to use (via the datasets library).'''} )
lowercase__: Optional[str] = field(
default=__lowercase , metadata={'''help''': '''The configuration name of the dataset to use (via the datasets library).'''} )
lowercase__: Optional[str] = field(
default='''train''' , metadata={
'''help''': '''The name of the training data set split to use (via the datasets library). Defaults to \'train\''''
} , )
lowercase__: Optional[str] = field(
default='''validation''' , metadata={
'''help''': (
'''The name of the validation data set split to use (via the datasets library). Defaults to \'validation\''''
)
} , )
lowercase__: Optional[str] = field(
default='''file''' , metadata={'''help''': '''Column in the dataset that contains speech file path. Defaults to \'file\''''} , )
lowercase__: bool = field(
default=__lowercase , metadata={'''help''': '''Overwrite the cached preprocessed datasets or not.'''} )
lowercase__: Optional[int] = field(
default=1 , metadata={
'''help''': '''The percentage of the train set used as validation set in case there\'s no validation split'''
} , )
lowercase__: Optional[int] = field(
default=__lowercase , metadata={'''help''': '''The number of processes to use for the preprocessing.'''} , )
lowercase__: Optional[float] = field(
default=2_0.0 , metadata={'''help''': '''Filter audio files that are longer than `max_duration_in_seconds` seconds'''} )
@dataclass
class _A :
lowercase__: WavaVecaForPreTraining
lowercase__: WavaVecaFeatureExtractor
lowercase__: Union[bool, str] = "longest"
lowercase__: Optional[int] = None
lowercase__: Optional[int] = None
def __call__( self : Any , __magic_name__ : List[Dict[str, Union[List[int], torch.Tensor]]] ) -> Dict[str, torch.Tensor]:
"""simple docstring"""
__snake_case : Any = self.feature_extractor.pad(
__magic_name__ , max_length=self.max_length , padding=self.padding , pad_to_multiple_of=self.pad_to_multiple_of , return_tensors="""pt""" , )
__snake_case : int = self.model._get_feat_extract_output_lengths(batch["""input_values"""].shape[-1] )
__snake_case : Optional[Any] = batch["""input_values"""].shape[0]
# make sure that no loss is computed on padded inputs
if batch["attention_mask"] is not None:
# compute real output lengths according to convolution formula
__snake_case : Optional[int] = self.model._get_feat_extract_output_lengths(batch["""attention_mask"""].sum(-1 ) ).to(
torch.long )
__snake_case : Any = torch.zeros(
(batch_size, mask_indices_seq_length) , dtype=torch.long , device=batch["""input_values"""].device )
# these two operations makes sure that all values
# before the output lengths indices are attended to
__snake_case : Dict = 1
__snake_case : Optional[Any] = attention_mask.flip([-1] ).cumsum(-1 ).flip([-1] ).bool()
# sample randomly masked indices
__snake_case : int = _compute_mask_indices(
(batch_size, mask_indices_seq_length) , self.model.config.mask_time_prob , self.model.config.mask_time_length , attention_mask=__magic_name__ , min_masks=2 , )
return batch
class _A ( __lowercase ):
def __init__( self : str , *__magic_name__ : Any , __magic_name__ : Optional[Any]=1 , __magic_name__ : int=0 , __magic_name__ : Tuple=1.0 , **__magic_name__ : Any ) -> Dict:
"""simple docstring"""
super().__init__(*__magic_name__ , **__magic_name__ )
__snake_case : List[str] = 0
__snake_case : Optional[Any] = max_gumbel_temp
__snake_case : int = min_gumbel_temp
__snake_case : Union[str, Any] = gumbel_temp_decay
def lowercase__ ( self : Optional[Any] , __magic_name__ : nn.Module , __magic_name__ : Dict[str, Union[torch.Tensor, Any]] ) -> torch.Tensor:
"""simple docstring"""
model.train()
__snake_case : int = self._prepare_inputs(__magic_name__ )
if self.use_amp:
with autocast():
__snake_case : str = self.compute_loss(__magic_name__ , __magic_name__ )
else:
__snake_case : Tuple = self.compute_loss(__magic_name__ , __magic_name__ )
if self.args.n_gpu > 1 or self.deepspeed:
if model.module.config.ctc_loss_reduction == "mean":
__snake_case : Union[str, Any] = loss.mean()
elif model.module.config.ctc_loss_reduction == "sum":
__snake_case : str = loss.sum() / (inputs["""mask_time_indices"""]).sum()
else:
raise ValueError(f'''{model.config.ctc_loss_reduction} is not valid. Choose one of [\'mean\', \'sum\']''' )
if self.args.gradient_accumulation_steps > 1:
__snake_case : List[Any] = loss / self.args.gradient_accumulation_steps
if self.use_amp:
self.scaler.scale(__magic_name__ ).backward()
elif self.use_apex:
with amp.scale_loss(__magic_name__ , self.optimizer ) as scaled_loss:
scaled_loss.backward()
elif self.deepspeed:
self.deepspeed.backward(__magic_name__ )
else:
loss.backward()
self.num_update_step += 1
# make sure gumbel softmax temperature is decayed
if self.args.n_gpu > 1 or self.deepspeed:
model.module.set_gumbel_temperature(
max(self.max_gumbel_temp * self.gumbel_temp_decay**self.num_update_step , self.min_gumbel_temp ) )
else:
model.set_gumbel_temperature(
max(self.max_gumbel_temp * self.gumbel_temp_decay**self.num_update_step , self.min_gumbel_temp ) )
return loss.detach()
def _a ( ) -> Any:
"""simple docstring"""
__snake_case : Union[str, Any] = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments) )
__snake_case , __snake_case , __snake_case : List[Any] = parser.parse_args_into_dataclasses()
configure_logger(_lowerCamelCase , _lowerCamelCase )
# Downloading and loading a dataset from the hub.
__snake_case : Tuple = load_dataset(data_args.dataset_name , data_args.dataset_config_name , cache_dir=model_args.cache_dir )
if "validation" not in datasets.keys():
# make sure only "validation" and "train" keys remain"
__snake_case : int = DatasetDict()
__snake_case : Union[str, Any] = load_dataset(
data_args.dataset_name , data_args.dataset_config_name , split=F'''{data_args.train_split_name}[:{data_args.validation_split_percentage}%]''' , cache_dir=model_args.cache_dir , )
__snake_case : Optional[int] = load_dataset(
data_args.dataset_name , data_args.dataset_config_name , split=F'''{data_args.train_split_name}[{data_args.validation_split_percentage}%:]''' , cache_dir=model_args.cache_dir , )
else:
# make sure only "validation" and "train" keys remain"
__snake_case : Optional[Any] = DatasetDict()
__snake_case : Optional[Any] = load_dataset(
data_args.dataset_name , data_args.dataset_config_name , split="""validation""" , cache_dir=model_args.cache_dir , )
__snake_case : str = load_dataset(
data_args.dataset_name , data_args.dataset_config_name , split=F'''{data_args.train_split_name}''' , cache_dir=model_args.cache_dir , )
# only normalized-inputs-training is supported
__snake_case : Union[str, Any] = WavaVecaFeatureExtractor.from_pretrained(
model_args.model_name_or_path , cache_dir=model_args.cache_dir , do_normalize=_lowerCamelCase )
def prepare_dataset(_lowerCamelCase ):
# check that all files have the correct sampling rate
__snake_case , __snake_case : int = librosa.load(batch[data_args.speech_file_column] , sr=feature_extractor.sampling_rate )
return batch
# load audio files into numpy arrays
__snake_case : List[str] = datasets.map(
_lowerCamelCase , num_proc=data_args.preprocessing_num_workers , remove_columns=datasets["""train"""].column_names )
# filter audio files that are too long
__snake_case : Optional[Any] = vectorized_datasets.filter(
lambda _lowerCamelCase : len(data["""speech"""] ) < int(data_args.max_duration_in_seconds * feature_extractor.sampling_rate ) )
def normalize(_lowerCamelCase ):
return feature_extractor(batch["""speech"""] , sampling_rate=feature_extractor.sampling_rate )
# normalize and transform to `BatchFeatures`
__snake_case : str = vectorized_datasets.map(
_lowerCamelCase , batched=_lowerCamelCase , num_proc=data_args.preprocessing_num_workers , load_from_cache_file=not data_args.overwrite_cache , remove_columns=vectorized_datasets["""train"""].column_names , )
# pretraining is only supported for "newer" stable layer norm architecture
# apply_spec_augment has to be True, mask_feature_prob has to be 0.0
__snake_case : str = WavaVecaConfig.from_pretrained(
model_args.model_name_or_path , cache_dir=model_args.cache_dir , gradient_checkpointing=training_args.gradient_checkpointing , )
if not config.do_stable_layer_norm or config.feat_extract_norm != "layer":
raise ValueError(
"""PreTraining is only supported for ``config.do_stable_layer_norm=True`` and"""
""" ``config.feat_extract_norm='layer'""" )
__snake_case : Tuple = WavaVecaForPreTraining(_lowerCamelCase )
__snake_case : int = DataCollatorForWavaVecaPretraining(model=_lowerCamelCase , feature_extractor=_lowerCamelCase )
__snake_case : List[str] = WavaVecaPreTrainer(
model=_lowerCamelCase , data_collator=_lowerCamelCase , args=_lowerCamelCase , train_dataset=vectorized_datasets["""train"""] , eval_dataset=vectorized_datasets["""validation"""] , tokenizer=_lowerCamelCase , max_gumbel_temp=model_args.max_gumbel_temperature , min_gumbel_temp=model_args.min_gumbel_temperature , gumbel_temp_decay=model_args.gumbel_temperature_decay , )
trainer.train()
if __name__ == "__main__":
main()
| 26 |
import math
import os
from copy import deepcopy
import datasets
import evaluate
import torch
import transformers
from datasets import load_dataset
from torch.utils.data import DataLoader
from transformers import AutoModelForSequenceClassification, AutoTokenizer
from accelerate import Accelerator
from accelerate.test_utils import RegressionDataset, RegressionModel
from accelerate.utils import is_tpu_available, set_seed
__a :Optional[Any] = 'true'
def __snake_case ( __UpperCamelCase : Any ,__UpperCamelCase : List[Any]=82 ,__UpperCamelCase : Dict=16 ):
"""simple docstring"""
set_seed(42 )
A_ = RegressionModel()
A_ = deepcopy(__UpperCamelCase )
A_ = RegressionDataset(length=__UpperCamelCase )
A_ = DataLoader(__UpperCamelCase ,batch_size=__UpperCamelCase )
model.to(accelerator.device )
A_ , A_ = accelerator.prepare(__UpperCamelCase ,__UpperCamelCase )
return model, ddp_model, dataloader
def __snake_case ( __UpperCamelCase : Accelerator ,__UpperCamelCase : Dict=False ):
"""simple docstring"""
A_ = AutoTokenizer.from_pretrained("hf-internal-testing/mrpc-bert-base-cased" )
A_ = load_dataset("glue" ,"mrpc" ,split="validation" )
def tokenize_function(__UpperCamelCase : Optional[Any] ):
A_ = tokenizer(examples["sentence1"] ,examples["sentence2"] ,truncation=__UpperCamelCase ,max_length=__UpperCamelCase )
return outputs
with accelerator.main_process_first():
A_ = dataset.map(
__UpperCamelCase ,batched=__UpperCamelCase ,remove_columns=["idx", "sentence1", "sentence2"] ,)
A_ = tokenized_datasets.rename_column("label" ,"labels" )
def collate_fn(__UpperCamelCase : Union[str, Any] ):
if use_longest:
return tokenizer.pad(__UpperCamelCase ,padding="longest" ,return_tensors="pt" )
return tokenizer.pad(__UpperCamelCase ,padding="max_length" ,max_length=128 ,return_tensors="pt" )
return DataLoader(__UpperCamelCase ,shuffle=__UpperCamelCase ,collate_fn=__UpperCamelCase ,batch_size=16 )
def __snake_case ( __UpperCamelCase : Optional[Any] ,__UpperCamelCase : str ):
"""simple docstring"""
A_ = Accelerator(dispatch_batches=__UpperCamelCase ,split_batches=__UpperCamelCase )
A_ = get_dataloader(__UpperCamelCase ,not dispatch_batches )
A_ = AutoModelForSequenceClassification.from_pretrained(
"hf-internal-testing/mrpc-bert-base-cased" ,return_dict=__UpperCamelCase )
A_ , A_ = accelerator.prepare(__UpperCamelCase ,__UpperCamelCase )
return {"ddp": [ddp_model, ddp_dataloader, "cuda:0"], "no": [model, dataloader, accelerator.device]}, accelerator
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : int ,__UpperCamelCase : Optional[Any] ):
"""simple docstring"""
A_ = []
for batch in dataloader:
A_ , A_ = batch.values()
with torch.no_grad():
A_ = model(__UpperCamelCase )
A_ , A_ = accelerator.gather_for_metrics((logit, target) )
logits_and_targets.append((logit, target) )
A_ , A_ = [], []
for logit, targ in logits_and_targets:
logits.append(__UpperCamelCase )
targs.append(__UpperCamelCase )
A_ , A_ = torch.cat(__UpperCamelCase ), torch.cat(__UpperCamelCase )
return logits, targs
def __snake_case ( __UpperCamelCase : Accelerator ,__UpperCamelCase : Dict=82 ,__UpperCamelCase : List[Any]=False ,__UpperCamelCase : Dict=False ,__UpperCamelCase : Optional[int]=16 ):
"""simple docstring"""
A_ , A_ , A_ = get_basic_setup(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
A_ , A_ = generate_predictions(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
assert (
len(__UpperCamelCase ) == num_samples
), f'''Unexpected number of inputs:\n Expected: {num_samples}\n Actual: {len(__UpperCamelCase )}'''
def __snake_case ( __UpperCamelCase : bool = False ,__UpperCamelCase : bool = False ):
"""simple docstring"""
A_ = evaluate.load("glue" ,"mrpc" )
A_ , A_ = get_mrpc_setup(__UpperCamelCase ,__UpperCamelCase )
# First do baseline
A_ , A_ , A_ = setup["no"]
model.to(__UpperCamelCase )
model.eval()
for batch in dataloader:
batch.to(__UpperCamelCase )
with torch.inference_mode():
A_ = model(**__UpperCamelCase )
A_ = outputs.logits.argmax(dim=-1 )
metric.add_batch(predictions=__UpperCamelCase ,references=batch["labels"] )
A_ = metric.compute()
# Then do distributed
A_ , A_ , A_ = setup["ddp"]
model.eval()
for batch in dataloader:
with torch.inference_mode():
A_ = model(**__UpperCamelCase )
A_ = outputs.logits.argmax(dim=-1 )
A_ = batch["labels"]
A_ , A_ = accelerator.gather_for_metrics((preds, references) )
metric.add_batch(predictions=__UpperCamelCase ,references=__UpperCamelCase )
A_ = metric.compute()
for key in "accuracy f1".split():
assert math.isclose(
baseline[key] ,distributed[key] ), f'''Baseline and Distributed are not the same for key {key}:\n\tBaseline: {baseline[key]}\n\tDistributed: {distributed[key]}\n'''
def __snake_case ( ):
"""simple docstring"""
A_ = Accelerator(split_batches=__UpperCamelCase ,dispatch_batches=__UpperCamelCase )
if accelerator.is_local_main_process:
datasets.utils.logging.set_verbosity_warning()
transformers.utils.logging.set_verbosity_warning()
else:
datasets.utils.logging.set_verbosity_error()
transformers.utils.logging.set_verbosity_error()
# These are a bit slower so they should only be ran on the GPU or TPU
if torch.cuda.is_available() or is_tpu_available():
if accelerator.is_local_main_process:
print("**Testing gather_for_metrics**" )
for split_batches in [True, False]:
for dispatch_batches in [True, False]:
if accelerator.is_local_main_process:
print(f'''With: `split_batches={split_batches}`, `dispatch_batches={dispatch_batches}`''' )
test_mrpc(__UpperCamelCase ,__UpperCamelCase )
accelerator.state._reset_state()
if accelerator.is_local_main_process:
print("**Test torch metrics**" )
for split_batches in [True, False]:
for dispatch_batches in [True, False]:
A_ = Accelerator(split_batches=__UpperCamelCase ,dispatch_batches=__UpperCamelCase )
if accelerator.is_local_main_process:
print(f'''With: `split_batches={split_batches}`, `dispatch_batches={dispatch_batches}`, length=99''' )
test_torch_metrics(__UpperCamelCase ,99 )
accelerator.state._reset_state()
if accelerator.is_local_main_process:
print("**Test last batch is not dropped when perfectly divisible**" )
A_ = Accelerator()
test_torch_metrics(__UpperCamelCase ,512 )
accelerator.state._reset_state()
def __snake_case ( __UpperCamelCase : List[Any] ):
"""simple docstring"""
main()
if __name__ == "__main__":
main() | 86 | 0 |
import gc
import unittest
from diffusers import FlaxStableDiffusionInpaintPipeline
from diffusers.utils import is_flax_available, load_image, slow
from diffusers.utils.testing_utils import require_flax
if is_flax_available():
import jax
import jax.numpy as jnp
from flax.jax_utils import replicate
from flax.training.common_utils import shard
@slow
@require_flax
class lowerCamelCase( unittest.TestCase ):
'''simple docstring'''
def lowerCAmelCase__ ( self ):
# clean up the VRAM after each test
super().tearDown()
gc.collect()
def lowerCAmelCase__ ( self ):
_A = load_image(
'https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main'
'/sd2-inpaint/init_image.png' )
_A = load_image(
'https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd2-inpaint/mask.png' )
_A = 'xvjiarui/stable-diffusion-2-inpainting'
_A, _A = FlaxStableDiffusionInpaintPipeline.from_pretrained(snake_case_ , safety_checker=snake_case_ )
_A = 'Face of a yellow cat, high resolution, sitting on a park bench'
_A = jax.random.PRNGKey(0 )
_A = 50
_A = jax.device_count()
_A = num_samples * [prompt]
_A = num_samples * [init_image]
_A = num_samples * [mask_image]
_A, _A, _A = pipeline.prepare_inputs(snake_case_ , snake_case_ , snake_case_ )
# shard inputs and rng
_A = replicate(snake_case_ )
_A = jax.random.split(snake_case_ , jax.device_count() )
_A = shard(snake_case_ )
_A = shard(snake_case_ )
_A = shard(snake_case_ )
_A = pipeline(
snake_case_ , snake_case_ , snake_case_ , snake_case_ , snake_case_ , snake_case_ , jit=snake_case_ )
_A = output.images.reshape(snake_case_ , 512 , 512 , 3 )
_A = images[0, 253:256, 253:256, -1]
_A = jnp.asarray(jax.device_get(image_slice.flatten() ) )
_A = jnp.array(
[0.361_1307, 0.3764_9736, 0.375_7408, 0.3821_3953, 0.3929_5167, 0.384_1631, 0.4155_4978, 0.413_7475, 0.421_7084] )
print(F"output_slice: {output_slice}" )
assert jnp.abs(output_slice - expected_slice ).max() < 1E-2
| 27 |
import argparse
import os
from transformers.utils import direct_transformers_import
# All paths are set with the intent you should run this script from the root of the repo with the command
# python utils/check_task_guides.py
__a :Optional[Any] = 'src/transformers'
__a :Tuple = 'docs/source/en/tasks'
def __snake_case ( __UpperCamelCase : List[Any] ,__UpperCamelCase : Dict ,__UpperCamelCase : int ):
"""simple docstring"""
with open(__UpperCamelCase ,"r" ,encoding="utf-8" ,newline="\n" ) as f:
A_ = f.readlines()
# Find the start prompt.
A_ = 0
while not lines[start_index].startswith(__UpperCamelCase ):
start_index += 1
start_index += 1
A_ = start_index
while not lines[end_index].startswith(__UpperCamelCase ):
end_index += 1
end_index -= 1
while len(lines[start_index] ) <= 1:
start_index += 1
while len(lines[end_index] ) <= 1:
end_index -= 1
end_index += 1
return "".join(lines[start_index:end_index] ), start_index, end_index, lines
# This is to make sure the transformers module imported is the one in the repo.
__a :List[str] = direct_transformers_import(TRANSFORMERS_PATH)
__a :Optional[Any] = {
'asr.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_CTC_MAPPING_NAMES,
'audio_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_AUDIO_CLASSIFICATION_MAPPING_NAMES,
'language_modeling.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_CAUSAL_LM_MAPPING_NAMES,
'image_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_IMAGE_CLASSIFICATION_MAPPING_NAMES,
'masked_language_modeling.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_MASKED_LM_MAPPING_NAMES,
'multiple_choice.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_MULTIPLE_CHOICE_MAPPING_NAMES,
'object_detection.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_OBJECT_DETECTION_MAPPING_NAMES,
'question_answering.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_QUESTION_ANSWERING_MAPPING_NAMES,
'semantic_segmentation.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEMANTIC_SEGMENTATION_MAPPING_NAMES,
'sequence_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING_NAMES,
'summarization.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES,
'token_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING_NAMES,
'translation.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES,
'video_classification.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_VIDEO_CLASSIFICATION_MAPPING_NAMES,
'document_question_answering.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_DOCUMENT_QUESTION_ANSWERING_MAPPING_NAMES,
'monocular_depth_estimation.md': transformers_module.models.auto.modeling_auto.MODEL_FOR_DEPTH_ESTIMATION_MAPPING_NAMES,
}
# This list contains model types used in some task guides that are not in `CONFIG_MAPPING_NAMES` (therefore not in any
# `MODEL_MAPPING_NAMES` or any `MODEL_FOR_XXX_MAPPING_NAMES`).
__a :Optional[Any] = {
'summarization.md': ('nllb',),
'translation.md': ('nllb',),
}
def __snake_case ( __UpperCamelCase : Tuple ):
"""simple docstring"""
A_ = TASK_GUIDE_TO_MODELS[task_guide]
A_ = SPECIAL_TASK_GUIDE_TO_MODEL_TYPES.get(__UpperCamelCase ,set() )
A_ = {
code: name
for code, name in transformers_module.MODEL_NAMES_MAPPING.items()
if (code in model_maping_names or code in special_model_types)
}
return ", ".join([f'''[{name}](../model_doc/{code})''' for code, name in model_names.items()] ) + "\n"
def __snake_case ( __UpperCamelCase : str ,__UpperCamelCase : List[str]=False ):
"""simple docstring"""
A_ , A_ , A_ , A_ = _find_text_in_file(
filename=os.path.join(__UpperCamelCase ,__UpperCamelCase ) ,start_prompt="<!--This tip is automatically generated by `make fix-copies`, do not fill manually!-->" ,end_prompt="<!--End of the generated tip-->" ,)
A_ = get_model_list_for_task(__UpperCamelCase )
if current_list != new_list:
if overwrite:
with open(os.path.join(__UpperCamelCase ,__UpperCamelCase ) ,"w" ,encoding="utf-8" ,newline="\n" ) as f:
f.writelines(lines[:start_index] + [new_list] + lines[end_index:] )
else:
raise ValueError(
f'''The list of models that can be used in the {task_guide} guide needs an update. Run `make fix-copies`'''
" to fix this." )
if __name__ == "__main__":
__a :int = argparse.ArgumentParser()
parser.add_argument('--fix_and_overwrite', action='store_true', help='Whether to fix inconsistencies.')
__a :Optional[Any] = parser.parse_args()
for task_guide in TASK_GUIDE_TO_MODELS.keys():
check_model_list_for_task(task_guide, args.fix_and_overwrite) | 86 | 0 |
'''simple docstring'''
from collections import Counter
from pathlib import Path
from typing import Optional, Tuple
import yaml
class _a ( yaml.SafeLoader ):
'''simple docstring'''
def UpperCamelCase_ ( self, A ):
'''simple docstring'''
SCREAMING_SNAKE_CASE : str = [self.constructed_objects[key_node] for key_node, _ in node.value]
SCREAMING_SNAKE_CASE : List[Any] = [tuple(A ) if isinstance(A, A ) else key for key in keys]
SCREAMING_SNAKE_CASE : List[str] = Counter(A )
SCREAMING_SNAKE_CASE : Tuple = [key for key in counter if counter[key] > 1]
if duplicate_keys:
raise TypeError(F"Got duplicate yaml keys: {duplicate_keys}" )
def UpperCamelCase_ ( self, A, A=False ):
'''simple docstring'''
SCREAMING_SNAKE_CASE : List[str] = super().construct_mapping(A, deep=A )
self._check_no_duplicates_on_constructed_node(A )
return mapping
def lowercase__( __UpperCamelCase: str ):
"""simple docstring"""
SCREAMING_SNAKE_CASE : Optional[int] = list(readme_content.splitlines() )
if full_content and full_content[0] == "---" and "---" in full_content[1:]:
SCREAMING_SNAKE_CASE : Any = full_content[1:].index('---' ) + 1
SCREAMING_SNAKE_CASE : Tuple = '\n'.join(full_content[1:sep_idx] )
return yamlblock, "\n".join(full_content[sep_idx + 1 :] )
return None, "\n".join(__UpperCamelCase )
class _a ( SCREAMING_SNAKE_CASE ):
'''simple docstring'''
A : Optional[Any] = {'''train_eval_index'''} # train-eval-index in the YAML metadata
@classmethod
def UpperCamelCase_ ( cls, A ):
'''simple docstring'''
with open(A, encoding='utf-8' ) as readme_file:
SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE : List[Any] = _split_yaml_from_readme(readme_file.read() )
if yaml_string is not None:
return cls.from_yaml_string(A )
else:
return cls()
def UpperCamelCase_ ( self, A ):
'''simple docstring'''
if path.exists():
with open(A, encoding='utf-8' ) as readme_file:
SCREAMING_SNAKE_CASE : Optional[Any] = readme_file.read()
else:
SCREAMING_SNAKE_CASE : Union[str, Any] = None
SCREAMING_SNAKE_CASE : List[str] = self._to_readme(A )
with open(A, 'w', encoding='utf-8' ) as readme_file:
readme_file.write(A )
def UpperCamelCase_ ( self, A = None ):
'''simple docstring'''
if readme_content is not None:
SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE : List[str] = _split_yaml_from_readme(A )
SCREAMING_SNAKE_CASE : Optional[int] = '---\n' + self.to_yaml_string() + '---\n' + content
else:
SCREAMING_SNAKE_CASE : Any = '---\n' + self.to_yaml_string() + '---\n'
return full_content
@classmethod
def UpperCamelCase_ ( cls, A ):
'''simple docstring'''
SCREAMING_SNAKE_CASE : int = yaml.load(A, Loader=_NoDuplicateSafeLoader ) or {}
# Convert the YAML keys to DatasetMetadata fields
SCREAMING_SNAKE_CASE : Union[str, Any] = {
(key.replace('-', '_' ) if key.replace('-', '_' ) in cls._FIELDS_WITH_DASHES else key): value
for key, value in metadata_dict.items()
}
return cls(**A )
def UpperCamelCase_ ( self ):
'''simple docstring'''
return yaml.safe_dump(
{
(key.replace('_', '-' ) if key in self._FIELDS_WITH_DASHES else key): value
for key, value in self.items()
}, sort_keys=A, allow_unicode=A, encoding='utf-8', ).decode('utf-8' )
UpperCamelCase_ = {
"image-classification": [],
"translation": [],
"image-segmentation": [],
"fill-mask": [],
"automatic-speech-recognition": [],
"token-classification": [],
"sentence-similarity": [],
"audio-classification": [],
"question-answering": [],
"summarization": [],
"zero-shot-classification": [],
"table-to-text": [],
"feature-extraction": [],
"other": [],
"multiple-choice": [],
"text-classification": [],
"text-to-image": [],
"text2text-generation": [],
"zero-shot-image-classification": [],
"tabular-classification": [],
"tabular-regression": [],
"image-to-image": [],
"tabular-to-text": [],
"unconditional-image-generation": [],
"text-retrieval": [],
"text-to-speech": [],
"object-detection": [],
"audio-to-audio": [],
"text-generation": [],
"conversational": [],
"table-question-answering": [],
"visual-question-answering": [],
"image-to-text": [],
"reinforcement-learning": [],
"voice-activity-detection": [],
"time-series-forecasting": [],
"document-question-answering": [],
}
if __name__ == "__main__":
from argparse import ArgumentParser
UpperCamelCase_ = ArgumentParser(usage="Validate the yaml metadata block of a README.md file.")
ap.add_argument("readme_filepath")
UpperCamelCase_ = ap.parse_args()
UpperCamelCase_ = Path(args.readme_filepath)
UpperCamelCase_ = DatasetMetadata.from_readme(readme_filepath)
print(dataset_metadata)
dataset_metadata.to_readme(readme_filepath)
| 28 |
import argparse
import json
from pathlib import Path
import requests
import timm
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from transformers import DeiTImageProcessor, ViTConfig, ViTForImageClassification, ViTImageProcessor, ViTModel
from transformers.utils import logging
logging.set_verbosity_info()
__a :Dict = logging.get_logger(__name__)
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : Tuple=False ):
"""simple docstring"""
A_ = []
for i in range(config.num_hidden_layers ):
# encoder layers: output projection, 2 feedforward neural networks and 2 layernorms
rename_keys.append((f'''blocks.{i}.norm1.weight''', f'''vit.encoder.layer.{i}.layernorm_before.weight''') )
rename_keys.append((f'''blocks.{i}.norm1.bias''', f'''vit.encoder.layer.{i}.layernorm_before.bias''') )
rename_keys.append((f'''blocks.{i}.attn.proj.weight''', f'''vit.encoder.layer.{i}.attention.output.dense.weight''') )
rename_keys.append((f'''blocks.{i}.attn.proj.bias''', f'''vit.encoder.layer.{i}.attention.output.dense.bias''') )
rename_keys.append((f'''blocks.{i}.norm2.weight''', f'''vit.encoder.layer.{i}.layernorm_after.weight''') )
rename_keys.append((f'''blocks.{i}.norm2.bias''', f'''vit.encoder.layer.{i}.layernorm_after.bias''') )
rename_keys.append((f'''blocks.{i}.mlp.fc1.weight''', f'''vit.encoder.layer.{i}.intermediate.dense.weight''') )
rename_keys.append((f'''blocks.{i}.mlp.fc1.bias''', f'''vit.encoder.layer.{i}.intermediate.dense.bias''') )
rename_keys.append((f'''blocks.{i}.mlp.fc2.weight''', f'''vit.encoder.layer.{i}.output.dense.weight''') )
rename_keys.append((f'''blocks.{i}.mlp.fc2.bias''', f'''vit.encoder.layer.{i}.output.dense.bias''') )
# projection layer + position embeddings
rename_keys.extend(
[
("cls_token", "vit.embeddings.cls_token"),
("patch_embed.proj.weight", "vit.embeddings.patch_embeddings.projection.weight"),
("patch_embed.proj.bias", "vit.embeddings.patch_embeddings.projection.bias"),
("pos_embed", "vit.embeddings.position_embeddings"),
] )
if base_model:
# layernorm + pooler
rename_keys.extend(
[
("norm.weight", "layernorm.weight"),
("norm.bias", "layernorm.bias"),
("pre_logits.fc.weight", "pooler.dense.weight"),
("pre_logits.fc.bias", "pooler.dense.bias"),
] )
# if just the base model, we should remove "vit" from all keys that start with "vit"
A_ = [(pair[0], pair[1][4:]) if pair[1].startswith("vit" ) else pair for pair in rename_keys]
else:
# layernorm + classification head
rename_keys.extend(
[
("norm.weight", "vit.layernorm.weight"),
("norm.bias", "vit.layernorm.bias"),
("head.weight", "classifier.weight"),
("head.bias", "classifier.bias"),
] )
return rename_keys
def __snake_case ( __UpperCamelCase : Optional[int] ,__UpperCamelCase : Tuple ,__UpperCamelCase : Any=False ):
"""simple docstring"""
for i in range(config.num_hidden_layers ):
if base_model:
A_ = ""
else:
A_ = "vit."
# read in weights + bias of input projection layer (in timm, this is a single matrix + bias)
A_ = state_dict.pop(f'''blocks.{i}.attn.qkv.weight''' )
A_ = state_dict.pop(f'''blocks.{i}.attn.qkv.bias''' )
# next, add query, keys and values (in that order) to the state dict
A_ = in_proj_weight[
: config.hidden_size, :
]
A_ = in_proj_bias[: config.hidden_size]
A_ = in_proj_weight[
config.hidden_size : config.hidden_size * 2, :
]
A_ = in_proj_bias[
config.hidden_size : config.hidden_size * 2
]
A_ = in_proj_weight[
-config.hidden_size :, :
]
A_ = in_proj_bias[-config.hidden_size :]
def __snake_case ( __UpperCamelCase : List[Any] ):
"""simple docstring"""
A_ = ["head.weight", "head.bias"]
for k in ignore_keys:
state_dict.pop(__UpperCamelCase ,__UpperCamelCase )
def __snake_case ( __UpperCamelCase : Any ,__UpperCamelCase : Optional[int] ,__UpperCamelCase : List[str] ):
"""simple docstring"""
A_ = dct.pop(__UpperCamelCase )
A_ = val
def __snake_case ( ):
"""simple docstring"""
A_ = "http://images.cocodataset.org/val2017/000000039769.jpg"
A_ = Image.open(requests.get(__UpperCamelCase ,stream=__UpperCamelCase ).raw )
return im
@torch.no_grad()
def __snake_case ( __UpperCamelCase : Union[str, Any] ,__UpperCamelCase : Optional[int] ):
"""simple docstring"""
A_ = ViTConfig()
A_ = False
# dataset (ImageNet-21k only or also fine-tuned on ImageNet 2012), patch_size and image_size
if vit_name[-5:] == "in21k":
A_ = True
A_ = int(vit_name[-12:-10] )
A_ = int(vit_name[-9:-6] )
else:
A_ = 1000
A_ = "huggingface/label-files"
A_ = "imagenet-1k-id2label.json"
A_ = json.load(open(hf_hub_download(__UpperCamelCase ,__UpperCamelCase ,repo_type="dataset" ) ,"r" ) )
A_ = {int(__UpperCamelCase ): v for k, v in idalabel.items()}
A_ = idalabel
A_ = {v: k for k, v in idalabel.items()}
A_ = int(vit_name[-6:-4] )
A_ = int(vit_name[-3:] )
# size of the architecture
if "deit" in vit_name:
if vit_name[9:].startswith("tiny" ):
A_ = 192
A_ = 768
A_ = 12
A_ = 3
elif vit_name[9:].startswith("small" ):
A_ = 384
A_ = 1536
A_ = 12
A_ = 6
else:
pass
else:
if vit_name[4:].startswith("small" ):
A_ = 768
A_ = 2304
A_ = 8
A_ = 8
elif vit_name[4:].startswith("base" ):
pass
elif vit_name[4:].startswith("large" ):
A_ = 1024
A_ = 4096
A_ = 24
A_ = 16
elif vit_name[4:].startswith("huge" ):
A_ = 1280
A_ = 5120
A_ = 32
A_ = 16
# load original model from timm
A_ = timm.create_model(__UpperCamelCase ,pretrained=__UpperCamelCase )
timm_model.eval()
# load state_dict of original model, remove and rename some keys
A_ = timm_model.state_dict()
if base_model:
remove_classification_head_(__UpperCamelCase )
A_ = create_rename_keys(__UpperCamelCase ,__UpperCamelCase )
for src, dest in rename_keys:
rename_key(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
read_in_q_k_v(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
# load HuggingFace model
if vit_name[-5:] == "in21k":
A_ = ViTModel(__UpperCamelCase ).eval()
else:
A_ = ViTForImageClassification(__UpperCamelCase ).eval()
model.load_state_dict(__UpperCamelCase )
# Check outputs on an image, prepared by ViTImageProcessor/DeiTImageProcessor
if "deit" in vit_name:
A_ = DeiTImageProcessor(size=config.image_size )
else:
A_ = ViTImageProcessor(size=config.image_size )
A_ = image_processor(images=prepare_img() ,return_tensors="pt" )
A_ = encoding["pixel_values"]
A_ = model(__UpperCamelCase )
if base_model:
A_ = timm_model.forward_features(__UpperCamelCase )
assert timm_pooled_output.shape == outputs.pooler_output.shape
assert torch.allclose(__UpperCamelCase ,outputs.pooler_output ,atol=1E-3 )
else:
A_ = timm_model(__UpperCamelCase )
assert timm_logits.shape == outputs.logits.shape
assert torch.allclose(__UpperCamelCase ,outputs.logits ,atol=1E-3 )
Path(__UpperCamelCase ).mkdir(exist_ok=__UpperCamelCase )
print(f'''Saving model {vit_name} to {pytorch_dump_folder_path}''' )
model.save_pretrained(__UpperCamelCase )
print(f'''Saving image processor to {pytorch_dump_folder_path}''' )
image_processor.save_pretrained(__UpperCamelCase )
if __name__ == "__main__":
__a :str = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
'--vit_name',
default='vit_base_patch16_224',
type=str,
help='Name of the ViT timm model you\'d like to convert.',
)
parser.add_argument(
'--pytorch_dump_folder_path', default=None, type=str, help='Path to the output PyTorch model directory.'
)
__a :Optional[int] = parser.parse_args()
convert_vit_checkpoint(args.vit_name, args.pytorch_dump_folder_path) | 86 | 0 |
"""simple docstring"""
import argparse
import json
from pathlib import Path
import requests
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from transformers import BeitConfig, BeitForImageClassification, BeitForMaskedImageModeling, BeitImageProcessor
from transformers.image_utils import PILImageResampling
from transformers.utils import logging
logging.set_verbosity_info()
A_ = logging.get_logger(__name__)
def lowercase ( lowerCAmelCase__ ,lowerCAmelCase__=False ,lowerCAmelCase__=False ):
lowerCamelCase_ = '''backbone.''' if is_semantic else ''''''
lowerCamelCase_ = []
for i in range(config.num_hidden_layers ):
# encoder layers: output projection, 2 feedforward neural networks and 2 layernorms
rename_keys.append((f"{prefix}blocks.{i}.norm1.weight", f"beit.encoder.layer.{i}.layernorm_before.weight") )
rename_keys.append((f"{prefix}blocks.{i}.norm1.bias", f"beit.encoder.layer.{i}.layernorm_before.bias") )
rename_keys.append(
(f"{prefix}blocks.{i}.attn.proj.weight", f"beit.encoder.layer.{i}.attention.output.dense.weight") )
rename_keys.append(
(f"{prefix}blocks.{i}.attn.proj.bias", f"beit.encoder.layer.{i}.attention.output.dense.bias") )
rename_keys.append((f"{prefix}blocks.{i}.norm2.weight", f"beit.encoder.layer.{i}.layernorm_after.weight") )
rename_keys.append((f"{prefix}blocks.{i}.norm2.bias", f"beit.encoder.layer.{i}.layernorm_after.bias") )
rename_keys.append((f"{prefix}blocks.{i}.mlp.fc1.weight", f"beit.encoder.layer.{i}.intermediate.dense.weight") )
rename_keys.append((f"{prefix}blocks.{i}.mlp.fc1.bias", f"beit.encoder.layer.{i}.intermediate.dense.bias") )
rename_keys.append((f"{prefix}blocks.{i}.mlp.fc2.weight", f"beit.encoder.layer.{i}.output.dense.weight") )
rename_keys.append((f"{prefix}blocks.{i}.mlp.fc2.bias", f"beit.encoder.layer.{i}.output.dense.bias") )
# projection layer + position embeddings
rename_keys.extend(
[
(f"{prefix}cls_token", '''beit.embeddings.cls_token'''),
(f"{prefix}patch_embed.proj.weight", '''beit.embeddings.patch_embeddings.projection.weight'''),
(f"{prefix}patch_embed.proj.bias", '''beit.embeddings.patch_embeddings.projection.bias'''),
(f"{prefix}pos_embed", '''beit.embeddings.position_embeddings'''),
] )
if has_lm_head:
# mask token + layernorm
rename_keys.extend(
[
('''mask_token''', '''beit.embeddings.mask_token'''),
('''norm.weight''', '''layernorm.weight'''),
('''norm.bias''', '''layernorm.bias'''),
] )
else:
# layernorm + classification head
rename_keys.extend(
[
('''fc_norm.weight''', '''beit.pooler.layernorm.weight'''),
('''fc_norm.bias''', '''beit.pooler.layernorm.bias'''),
('''head.weight''', '''classifier.weight'''),
('''head.bias''', '''classifier.bias'''),
] )
return rename_keys
def lowercase ( lowerCAmelCase__ ,lowerCAmelCase__ ,lowerCAmelCase__=False ,lowerCAmelCase__=False ):
for i in range(config.num_hidden_layers ):
lowerCamelCase_ = '''backbone.''' if is_semantic else ''''''
# queries, keys and values
lowerCamelCase_ = state_dict.pop(f"{prefix}blocks.{i}.attn.qkv.weight" )
lowerCamelCase_ = state_dict.pop(f"{prefix}blocks.{i}.attn.q_bias" )
lowerCamelCase_ = state_dict.pop(f"{prefix}blocks.{i}.attn.v_bias" )
lowerCamelCase_ = in_proj_weight[
: config.hidden_size, :
]
lowerCamelCase_ = q_bias
lowerCamelCase_ = in_proj_weight[
config.hidden_size : config.hidden_size * 2, :
]
lowerCamelCase_ = in_proj_weight[
-config.hidden_size :, :
]
lowerCamelCase_ = v_bias
# gamma_1 and gamma_2
# we call them lambda because otherwise they are renamed when using .from_pretrained
lowerCamelCase_ = state_dict.pop(f"{prefix}blocks.{i}.gamma_1" )
lowerCamelCase_ = state_dict.pop(f"{prefix}blocks.{i}.gamma_2" )
lowerCamelCase_ = gamma_a
lowerCamelCase_ = gamma_a
def lowercase ( lowerCAmelCase__ ,lowerCAmelCase__ ,lowerCAmelCase__ ):
lowerCamelCase_ = dct.pop(lowerCAmelCase__ )
lowerCamelCase_ = val
def lowercase ( ):
lowerCamelCase_ = '''http://images.cocodataset.org/val2017/000000039769.jpg'''
lowerCamelCase_ = Image.open(requests.get(lowerCAmelCase__ ,stream=lowerCAmelCase__ ).raw )
return im
@torch.no_grad()
def lowercase ( lowerCAmelCase__ ,lowerCAmelCase__ ,lowerCAmelCase__=False ):
lowerCamelCase_ = False if '''rvlcdip''' in checkpoint_url else True
lowerCamelCase_ = BeitConfig(use_absolute_position_embeddings=lowerCAmelCase__ ,use_mask_token=lowerCAmelCase__ )
# size of the architecture
if "large" in checkpoint_url or "dit-l" in checkpoint_url:
lowerCamelCase_ = 1_024
lowerCamelCase_ = 4_096
lowerCamelCase_ = 24
lowerCamelCase_ = 16
# labels
if "rvlcdip" in checkpoint_url:
lowerCamelCase_ = 16
lowerCamelCase_ = '''huggingface/label-files'''
lowerCamelCase_ = '''rvlcdip-id2label.json'''
lowerCamelCase_ = json.load(open(hf_hub_download(lowerCAmelCase__ ,lowerCAmelCase__ ,repo_type='''dataset''' ) ,'''r''' ) )
lowerCamelCase_ = {int(lowerCAmelCase__ ): v for k, v in idalabel.items()}
lowerCamelCase_ = idalabel
lowerCamelCase_ = {v: k for k, v in idalabel.items()}
# load state_dict of original model, remove and rename some keys
lowerCamelCase_ = torch.hub.load_state_dict_from_url(lowerCAmelCase__ ,map_location='''cpu''' )['''model''']
lowerCamelCase_ = create_rename_keys(lowerCAmelCase__ ,has_lm_head=lowerCAmelCase__ )
for src, dest in rename_keys:
rename_key(lowerCAmelCase__ ,lowerCAmelCase__ ,lowerCAmelCase__ )
read_in_q_k_v(lowerCAmelCase__ ,lowerCAmelCase__ ,has_lm_head=lowerCAmelCase__ )
# load HuggingFace model
lowerCamelCase_ = BeitForMaskedImageModeling(lowerCAmelCase__ ) if has_lm_head else BeitForImageClassification(lowerCAmelCase__ )
model.eval()
model.load_state_dict(lowerCAmelCase__ )
# Check outputs on an image
lowerCamelCase_ = BeitImageProcessor(
size=config.image_size ,resample=PILImageResampling.BILINEAR ,do_center_crop=lowerCAmelCase__ )
lowerCamelCase_ = prepare_img()
lowerCamelCase_ = image_processor(images=lowerCAmelCase__ ,return_tensors='''pt''' )
lowerCamelCase_ = encoding['''pixel_values''']
lowerCamelCase_ = model(lowerCAmelCase__ )
lowerCamelCase_ = outputs.logits
# verify logits
lowerCamelCase_ = [1, 16] if '''rvlcdip''' in checkpoint_url else [1, 196, 8_192]
assert logits.shape == torch.Size(lowerCAmelCase__ ), "Shape of logits not as expected"
Path(lowerCAmelCase__ ).mkdir(exist_ok=lowerCAmelCase__ )
print(f"Saving model to {pytorch_dump_folder_path}" )
model.save_pretrained(lowerCAmelCase__ )
print(f"Saving image processor to {pytorch_dump_folder_path}" )
image_processor.save_pretrained(lowerCAmelCase__ )
if push_to_hub:
if has_lm_head:
lowerCamelCase_ = '''dit-base''' if '''base''' in checkpoint_url else '''dit-large'''
else:
lowerCamelCase_ = '''dit-base-finetuned-rvlcdip''' if '''dit-b''' in checkpoint_url else '''dit-large-finetuned-rvlcdip'''
image_processor.push_to_hub(
repo_path_or_name=Path(lowerCAmelCase__ ,lowerCAmelCase__ ) ,organization='''nielsr''' ,commit_message='''Add image processor''' ,use_temp_dir=lowerCAmelCase__ ,)
model.push_to_hub(
repo_path_or_name=Path(lowerCAmelCase__ ,lowerCAmelCase__ ) ,organization='''nielsr''' ,commit_message='''Add model''' ,use_temp_dir=lowerCAmelCase__ ,)
if __name__ == "__main__":
A_ = argparse.ArgumentParser()
parser.add_argument(
"""--checkpoint_url""",
default="""https://layoutlm.blob.core.windows.net/dit/dit-pts/dit-base-224-p16-500k-62d53a.pth""",
type=str,
help="""URL to the original PyTorch checkpoint (.pth file).""",
)
parser.add_argument(
"""--pytorch_dump_folder_path""", default=None, type=str, help="""Path to the folder to output PyTorch model."""
)
parser.add_argument(
"""--push_to_hub""",
action="""store_true""",
)
A_ = parser.parse_args()
convert_dit_checkpoint(args.checkpoint_url, args.pytorch_dump_folder_path, args.push_to_hub)
| 29 |
def __snake_case ( __UpperCamelCase : int = 50 ):
"""simple docstring"""
A_ = [1] * (length + 1)
for row_length in range(length + 1 ):
for tile_length in range(2 ,5 ):
for tile_start in range(row_length - tile_length + 1 ):
ways_number[row_length] += ways_number[
row_length - tile_start - tile_length
]
return ways_number[length]
if __name__ == "__main__":
print(F"{solution() = }") | 86 | 0 |
import os
from shutil import copyfile
from typing import Any, Dict, List, Optional, Tuple
import sentencepiece as spm
from ...tokenization_utils import AddedToken, PreTrainedTokenizer
from ...utils import logging
__a = logging.get_logger(__name__)
__a = '▁'
__a = {'vocab_file': 'sentencepiece.bpe.model', 'monolingual_vocab_file': 'dict.txt'}
__a = {
'vocab_file': {
'vinai/bartpho-syllable': 'https://huggingface.co/vinai/bartpho-syllable/resolve/main/sentencepiece.bpe.model',
},
'monolingual_vocab_file': {
'vinai/bartpho-syllable': 'https://huggingface.co/vinai/bartpho-syllable/resolve/main/dict.txt',
},
}
__a = {'vinai/bartpho-syllable': 1_024}
class __a( _a ):
"""simple docstring"""
lowerCAmelCase = VOCAB_FILES_NAMES
lowerCAmelCase = PRETRAINED_VOCAB_FILES_MAP
lowerCAmelCase = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
lowerCAmelCase = ['''input_ids''', '''attention_mask''']
def __init__( self ,_SCREAMING_SNAKE_CASE ,_SCREAMING_SNAKE_CASE ,_SCREAMING_SNAKE_CASE="<s>" ,_SCREAMING_SNAKE_CASE="</s>" ,_SCREAMING_SNAKE_CASE="</s>" ,_SCREAMING_SNAKE_CASE="<s>" ,_SCREAMING_SNAKE_CASE="<unk>" ,_SCREAMING_SNAKE_CASE="<pad>" ,_SCREAMING_SNAKE_CASE="<mask>" ,_SCREAMING_SNAKE_CASE = None ,**_SCREAMING_SNAKE_CASE ,) -> None:
# Mask token behave like a normal word, i.e. include the space before it
UpperCAmelCase_ : List[Any] = AddedToken(_SCREAMING_SNAKE_CASE ,lstrip=_SCREAMING_SNAKE_CASE ,rstrip=_SCREAMING_SNAKE_CASE ) if isinstance(_SCREAMING_SNAKE_CASE ,_SCREAMING_SNAKE_CASE ) else mask_token
UpperCAmelCase_ : Any = {} if sp_model_kwargs is None else sp_model_kwargs
super().__init__(
bos_token=_SCREAMING_SNAKE_CASE ,eos_token=_SCREAMING_SNAKE_CASE ,unk_token=_SCREAMING_SNAKE_CASE ,sep_token=_SCREAMING_SNAKE_CASE ,cls_token=_SCREAMING_SNAKE_CASE ,pad_token=_SCREAMING_SNAKE_CASE ,mask_token=_SCREAMING_SNAKE_CASE ,sp_model_kwargs=self.sp_model_kwargs ,**_SCREAMING_SNAKE_CASE ,)
UpperCAmelCase_ : str = vocab_file
UpperCAmelCase_ : Any = monolingual_vocab_file
UpperCAmelCase_ : int = spm.SentencePieceProcessor(**self.sp_model_kwargs )
self.sp_model.Load(str(_SCREAMING_SNAKE_CASE ) )
# Load the reduced vocab
# Keep order of special tokens for backward compatibility
UpperCAmelCase_ : str = {}
UpperCAmelCase_ : Any = 0
for token in [bos_token, pad_token, eos_token, unk_token, sep_token, cls_token]:
if str(_SCREAMING_SNAKE_CASE ) not in self.fairseq_tokens_to_ids:
UpperCAmelCase_ : str = cnt
cnt += 1
with open(_SCREAMING_SNAKE_CASE ,'''r''' ,encoding='''utf-8''' ) as f:
for line in f.readlines():
UpperCAmelCase_ : int = line.strip().split()[0]
UpperCAmelCase_ : Tuple = len(self.fairseq_tokens_to_ids )
if str(_SCREAMING_SNAKE_CASE ) not in self.fairseq_tokens_to_ids:
UpperCAmelCase_ : Tuple = len(self.fairseq_tokens_to_ids )
UpperCAmelCase_ : Union[str, Any] = {v: k for k, v in self.fairseq_tokens_to_ids.items()}
def __getstate__( self ) -> str:
UpperCAmelCase_ : Any = self.__dict__.copy()
UpperCAmelCase_ : Optional[Any] = None
UpperCAmelCase_ : Tuple = self.sp_model.serialized_model_proto()
return state
def __setstate__( self ,_SCREAMING_SNAKE_CASE ) -> Optional[int]:
UpperCAmelCase_ : List[str] = d
# for backward compatibility
if not hasattr(self ,'''sp_model_kwargs''' ):
UpperCAmelCase_ : Optional[Any] = {}
UpperCAmelCase_ : str = spm.SentencePieceProcessor(**self.sp_model_kwargs )
self.sp_model.LoadFromSerializedProto(self.sp_model_proto )
def a__ ( self ,_SCREAMING_SNAKE_CASE ,_SCREAMING_SNAKE_CASE = None ) -> List[int]:
if token_ids_a is None:
return [self.cls_token_id] + token_ids_a + [self.sep_token_id]
UpperCAmelCase_ : Any = [self.cls_token_id]
UpperCAmelCase_ : Optional[Any] = [self.sep_token_id]
return cls + token_ids_a + sep + sep + token_ids_a + sep
def a__ ( self ,_SCREAMING_SNAKE_CASE ,_SCREAMING_SNAKE_CASE = None ,_SCREAMING_SNAKE_CASE = False ) -> List[int]:
if already_has_special_tokens:
return super().get_special_tokens_mask(
token_ids_a=_SCREAMING_SNAKE_CASE ,token_ids_a=_SCREAMING_SNAKE_CASE ,already_has_special_tokens=_SCREAMING_SNAKE_CASE )
if token_ids_a is None:
return [1] + ([0] * len(_SCREAMING_SNAKE_CASE )) + [1]
return [1] + ([0] * len(_SCREAMING_SNAKE_CASE )) + [1, 1] + ([0] * len(_SCREAMING_SNAKE_CASE )) + [1]
def a__ ( self ,_SCREAMING_SNAKE_CASE ,_SCREAMING_SNAKE_CASE = None ) -> List[int]:
UpperCAmelCase_ : Dict = [self.sep_token_id]
UpperCAmelCase_ : Dict = [self.cls_token_id]
if token_ids_a is None:
return len(cls + token_ids_a + sep ) * [0]
return len(cls + token_ids_a + sep + sep + token_ids_a + sep ) * [0]
@property
def a__ ( self ) -> Optional[Any]:
return len(self.fairseq_ids_to_tokens )
def a__ ( self ) -> List[str]:
UpperCAmelCase_ : List[str] = {self.convert_ids_to_tokens(_SCREAMING_SNAKE_CASE ): i for i in range(self.vocab_size )}
vocab.update(self.added_tokens_encoder )
return vocab
def a__ ( self ,_SCREAMING_SNAKE_CASE ) -> List[str]:
return self.sp_model.encode(_SCREAMING_SNAKE_CASE ,out_type=_SCREAMING_SNAKE_CASE )
def a__ ( self ,_SCREAMING_SNAKE_CASE ) -> Optional[int]:
if token in self.fairseq_tokens_to_ids:
return self.fairseq_tokens_to_ids[token]
else:
return self.unk_token_id
def a__ ( self ,_SCREAMING_SNAKE_CASE ) -> int:
return self.fairseq_ids_to_tokens[index]
def a__ ( self ,_SCREAMING_SNAKE_CASE ) -> Any:
UpperCAmelCase_ : Union[str, Any] = ''''''.join(_SCREAMING_SNAKE_CASE ).replace(_SCREAMING_SNAKE_CASE ,''' ''' ).strip()
return out_string
def a__ ( self ,_SCREAMING_SNAKE_CASE ,_SCREAMING_SNAKE_CASE = None ) -> Tuple[str]:
if not os.path.isdir(_SCREAMING_SNAKE_CASE ):
logger.error(f'''Vocabulary path ({save_directory}) should be a directory''' )
return
UpperCAmelCase_ : Tuple = os.path.join(
_SCREAMING_SNAKE_CASE ,(filename_prefix + '''-''' if filename_prefix else '''''') + VOCAB_FILES_NAMES['''vocab_file'''] )
UpperCAmelCase_ : List[str] = os.path.join(
_SCREAMING_SNAKE_CASE ,(filename_prefix + '''-''' if filename_prefix else '''''') + VOCAB_FILES_NAMES['''monolingual_vocab_file'''] ,)
if os.path.abspath(self.vocab_file ) != os.path.abspath(_SCREAMING_SNAKE_CASE ) and os.path.isfile(self.vocab_file ):
copyfile(self.vocab_file ,_SCREAMING_SNAKE_CASE )
elif not os.path.isfile(self.vocab_file ):
with open(_SCREAMING_SNAKE_CASE ,'''wb''' ) as fi:
UpperCAmelCase_ : str = self.sp_model.serialized_model_proto()
fi.write(_SCREAMING_SNAKE_CASE )
if os.path.abspath(self.monolingual_vocab_file ) != os.path.abspath(
_SCREAMING_SNAKE_CASE ) and os.path.isfile(self.monolingual_vocab_file ):
copyfile(self.monolingual_vocab_file ,_SCREAMING_SNAKE_CASE )
elif not os.path.isfile(self.monolingual_vocab_file ):
with open(_SCREAMING_SNAKE_CASE ,'''w''' ,encoding='''utf-8''' ) as fp:
for token in self.fairseq_tokens_to_ids:
if token not in self.all_special_tokens:
fp.write(f'''{str(_SCREAMING_SNAKE_CASE )} \n''' )
return out_vocab_file, out_monolingual_vocab_file | 30 |
from typing import Any, Dict, List, Union
from ..utils import add_end_docstrings, is_torch_available, is_vision_available, logging, requires_backends
from .base import PIPELINE_INIT_ARGS, ChunkPipeline
if is_vision_available():
from PIL import Image
from ..image_utils import load_image
if is_torch_available():
import torch
from transformers.modeling_outputs import BaseModelOutput
from ..models.auto.modeling_auto import MODEL_FOR_ZERO_SHOT_OBJECT_DETECTION_MAPPING
__a :List[str] = logging.get_logger(__name__)
@add_end_docstrings(snake_case_ )
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Any , **UpperCAmelCase : List[str] ):
super().__init__(**UpperCAmelCase )
if self.framework == "tf":
raise ValueError(f'''The {self.__class__} is only available in PyTorch.''' )
requires_backends(self , "vision" )
self.check_model_type(UpperCAmelCase )
def __call__( self : Optional[int] , UpperCAmelCase : Union[str, "Image.Image", List[Dict[str, Any]]] , UpperCAmelCase : Union[str, List[str]] = None , **UpperCAmelCase : List[Any] , ):
if "text_queries" in kwargs:
A_ = kwargs.pop("text_queries" )
if isinstance(UpperCAmelCase , (str, Image.Image) ):
A_ = {"image": image, "candidate_labels": candidate_labels}
else:
A_ = image
A_ = super().__call__(UpperCAmelCase , **UpperCAmelCase )
return results
def __A ( self : int , **UpperCAmelCase : Tuple ):
A_ = {}
if "threshold" in kwargs:
A_ = kwargs["threshold"]
if "top_k" in kwargs:
A_ = kwargs["top_k"]
return {}, {}, postprocess_params
def __A ( self : List[str] , UpperCAmelCase : Dict ):
A_ = load_image(inputs["image"] )
A_ = inputs["candidate_labels"]
if isinstance(UpperCAmelCase , UpperCAmelCase ):
A_ = candidate_labels.split("," )
A_ = torch.tensor([[image.height, image.width]] , dtype=torch.intaa )
for i, candidate_label in enumerate(UpperCAmelCase ):
A_ = self.tokenizer(UpperCAmelCase , return_tensors=self.framework )
A_ = self.image_processor(UpperCAmelCase , return_tensors=self.framework )
yield {
"is_last": i == len(UpperCAmelCase ) - 1,
"target_size": target_size,
"candidate_label": candidate_label,
**text_inputs,
**image_features,
}
def __A ( self : str , UpperCAmelCase : int ):
A_ = model_inputs.pop("target_size" )
A_ = model_inputs.pop("candidate_label" )
A_ = model_inputs.pop("is_last" )
A_ = self.model(**UpperCAmelCase )
A_ = {"target_size": target_size, "candidate_label": candidate_label, "is_last": is_last, **outputs}
return model_outputs
def __A ( self : Dict , UpperCAmelCase : Any , UpperCAmelCase : Optional[int]=0.1 , UpperCAmelCase : Optional[int]=None ):
A_ = []
for model_output in model_outputs:
A_ = model_output["candidate_label"]
A_ = BaseModelOutput(UpperCAmelCase )
A_ = self.image_processor.post_process_object_detection(
outputs=UpperCAmelCase , threshold=UpperCAmelCase , target_sizes=model_output["target_size"] )[0]
for index in outputs["scores"].nonzero():
A_ = outputs["scores"][index].item()
A_ = self._get_bounding_box(outputs["boxes"][index][0] )
A_ = {"score": score, "label": label, "box": box}
results.append(UpperCAmelCase )
A_ = sorted(UpperCAmelCase , key=lambda UpperCAmelCase : x["score"] , reverse=UpperCAmelCase )
if top_k:
A_ = results[:top_k]
return results
def __A ( self : List[str] , UpperCAmelCase : "torch.Tensor" ):
if self.framework != "pt":
raise ValueError("The ZeroShotObjectDetectionPipeline is only available in PyTorch." )
A_ , A_ , A_ , A_ = box.int().tolist()
A_ = {
"xmin": xmin,
"ymin": ymin,
"xmax": xmax,
"ymax": ymax,
}
return bbox | 86 | 0 |
import fire
from utils import calculate_rouge, save_json
def UpperCAmelCase_ ( __UpperCAmelCase : Dict , __UpperCAmelCase : List[Any] , __UpperCAmelCase : Tuple=None , **__UpperCAmelCase : List[str] ) -> Any:
SCREAMING_SNAKE_CASE_ = [x.strip() for x in open(__UpperCAmelCase ).readlines()]
SCREAMING_SNAKE_CASE_ = [x.strip() for x in open(__UpperCAmelCase ).readlines()][: len(__UpperCAmelCase )]
SCREAMING_SNAKE_CASE_ = calculate_rouge(__UpperCAmelCase , __UpperCAmelCase , **__UpperCAmelCase )
if save_path is not None:
save_json(__UpperCAmelCase , __UpperCAmelCase , indent=__UpperCAmelCase )
return metrics # these print nicely
if __name__ == "__main__":
fire.Fire(calculate_rouge_path) | 31 |
import argparse
import torch
# Step 1. clone https://github.com/microsoft/unilm
# Step 2. git checkout to https://github.com/microsoft/unilm/commit/b94ec76c36f02fb2b0bf0dcb0b8554a2185173cd
# Step 3. cd unilm
# Step 4. ln -s $(realpath wavlm/modules.py) ./ # create simlink
# import classes
from unilm.wavlm.WavLM import WavLM as WavLMOrig
from unilm.wavlm.WavLM import WavLMConfig as WavLMConfigOrig
from transformers import WavLMConfig, WavLMModel, logging
logging.set_verbosity_info()
__a :Any = logging.get_logger(__name__)
__a :int = {
'post_extract_proj': 'feature_projection.projection',
'encoder.pos_conv.0': 'encoder.pos_conv_embed.conv',
'self_attn.k_proj': 'encoder.layers.*.attention.k_proj',
'self_attn.v_proj': 'encoder.layers.*.attention.v_proj',
'self_attn.q_proj': 'encoder.layers.*.attention.q_proj',
'self_attn.out_proj': 'encoder.layers.*.attention.out_proj',
'self_attn.grep_linear': 'encoder.layers.*.attention.gru_rel_pos_linear',
'self_attn.relative_attention_bias': 'encoder.layers.*.attention.rel_attn_embed',
'self_attn.grep_a': 'encoder.layers.*.attention.gru_rel_pos_const',
'self_attn_layer_norm': 'encoder.layers.*.layer_norm',
'fc1': 'encoder.layers.*.feed_forward.intermediate_dense',
'fc2': 'encoder.layers.*.feed_forward.output_dense',
'final_layer_norm': 'encoder.layers.*.final_layer_norm',
'encoder.layer_norm': 'encoder.layer_norm',
'w2v_model.layer_norm': 'feature_projection.layer_norm',
'quantizer.weight_proj': 'quantizer.weight_proj',
'quantizer.vars': 'quantizer.codevectors',
'project_q': 'project_q',
'final_proj': 'project_hid',
'w2v_encoder.proj': 'ctc_proj',
'mask_emb': 'masked_spec_embed',
}
__a :Tuple = [
'ctc_proj',
'quantizer.weight_proj',
'quantizer.codevectors',
'project_q',
'project_hid',
]
def __snake_case ( __UpperCamelCase : List[str] ,__UpperCamelCase : List[Any] ,__UpperCamelCase : Dict ,__UpperCamelCase : Any ,__UpperCamelCase : Optional[int] ):
"""simple docstring"""
for attribute in key.split("." ):
A_ = getattr(__UpperCamelCase ,__UpperCamelCase )
if weight_type is not None:
A_ = getattr(__UpperCamelCase ,__UpperCamelCase ).shape
else:
A_ = hf_pointer.shape
assert hf_shape == value.shape, (
f'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
f''' {value.shape} for {full_name}'''
)
if weight_type == "weight":
A_ = value
elif weight_type == "weight_g":
A_ = value
elif weight_type == "weight_v":
A_ = value
elif weight_type == "bias":
A_ = value
else:
A_ = value
logger.info(f'''{key + "." + weight_type if weight_type is not None else ""} was initialized from {full_name}.''' )
def __snake_case ( __UpperCamelCase : Optional[int] ,__UpperCamelCase : Optional[Any] ):
"""simple docstring"""
A_ = []
A_ = fairseq_model.state_dict()
A_ = hf_model.feature_extractor
for name, value in fairseq_dict.items():
A_ = False
if "conv_layers" in name:
load_conv_layer(
__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,hf_model.config.feat_extract_norm == "group" ,)
A_ = True
else:
for key, mapped_key in MAPPING.items():
if key in name or key.split("w2v_model." )[-1] == name.split("." )[0]:
A_ = True
if "*" in mapped_key:
A_ = name.split(__UpperCamelCase )[0].split("." )[-2]
A_ = mapped_key.replace("*" ,__UpperCamelCase )
if "weight_g" in name:
A_ = "weight_g"
elif "weight_v" in name:
A_ = "weight_v"
elif "bias" in name and "relative_attention_bias" not in name:
A_ = "bias"
elif "weight" in name:
# TODO: don't match quantizer.weight_proj
A_ = "weight"
else:
A_ = None
set_recursively(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
continue
if not is_used:
unused_weights.append(__UpperCamelCase )
logger.warning(f'''Unused weights: {unused_weights}''' )
def __snake_case ( __UpperCamelCase : Optional[int] ,__UpperCamelCase : Dict ,__UpperCamelCase : List[str] ,__UpperCamelCase : Union[str, Any] ,__UpperCamelCase : Optional[int] ):
"""simple docstring"""
A_ = full_name.split("conv_layers." )[-1]
A_ = name.split("." )
A_ = int(items[0] )
A_ = int(items[1] )
if type_id == 0:
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.bias.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape, (
f'''{full_name} has size {value.shape}, but {feature_extractor[layer_id].layer_norm.bias.data.shape} was'''
" found."
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor[layer_id].layer_norm.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(__UpperCamelCase )
@torch.no_grad()
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : str ,__UpperCamelCase : int=None ):
"""simple docstring"""
A_ = torch.load(__UpperCamelCase )
A_ = WavLMConfigOrig(checkpoint["cfg"] )
A_ = WavLMOrig(__UpperCamelCase )
model.load_state_dict(checkpoint["model"] )
model.eval()
if config_path is not None:
A_ = WavLMConfig.from_pretrained(__UpperCamelCase )
else:
A_ = WavLMConfig()
A_ = WavLMModel(__UpperCamelCase )
recursively_load_weights(__UpperCamelCase ,__UpperCamelCase )
hf_wavlm.save_pretrained(__UpperCamelCase )
if __name__ == "__main__":
__a :List[Any] = argparse.ArgumentParser()
parser.add_argument('--pytorch_dump_folder_path', default=None, type=str, help='Path to the output PyTorch model.')
parser.add_argument('--checkpoint_path', default=None, type=str, help='Path to fairseq checkpoint')
parser.add_argument('--config_path', default=None, type=str, help='Path to hf config.json of model to convert')
__a :Optional[int] = parser.parse_args()
convert_wavlm_checkpoint(args.checkpoint_path, args.pytorch_dump_folder_path, args.config_path) | 86 | 0 |
import collections
import inspect
import unittest
from transformers import FocalNetConfig
from transformers.testing_utils import require_torch, require_vision, slow, torch_device
from transformers.utils import cached_property, is_torch_available, is_vision_available
from ...test_backbone_common import BackboneTesterMixin
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, _config_zero_init, floats_tensor, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from torch import nn
from transformers import (
FocalNetBackbone,
FocalNetForImageClassification,
FocalNetForMaskedImageModeling,
FocalNetModel,
)
from transformers.models.focalnet.modeling_focalnet import FOCALNET_PRETRAINED_MODEL_ARCHIVE_LIST
if is_vision_available():
from PIL import Image
from transformers import AutoImageProcessor
class __UpperCamelCase :
def __init__( self , _UpperCamelCase , _UpperCamelCase=13 , _UpperCamelCase=32 , _UpperCamelCase=2 , _UpperCamelCase=3 , _UpperCamelCase=16 , _UpperCamelCase=[32, 64, 128] , _UpperCamelCase=[1, 2, 1] , _UpperCamelCase=[2, 2, 4] , _UpperCamelCase=2 , _UpperCamelCase=2.0 , _UpperCamelCase=True , _UpperCamelCase=0.0 , _UpperCamelCase=0.0 , _UpperCamelCase=0.1 , _UpperCamelCase="gelu" , _UpperCamelCase=False , _UpperCamelCase=True , _UpperCamelCase=0.02 , _UpperCamelCase=1e-5 , _UpperCamelCase=True , _UpperCamelCase=None , _UpperCamelCase=True , _UpperCamelCase=10 , _UpperCamelCase=8 , _UpperCamelCase=["stage1", "stage2"] , _UpperCamelCase=[1, 2] , ):
_UpperCAmelCase = parent
_UpperCAmelCase = batch_size
_UpperCAmelCase = image_size
_UpperCAmelCase = patch_size
_UpperCAmelCase = num_channels
_UpperCAmelCase = embed_dim
_UpperCAmelCase = hidden_sizes
_UpperCAmelCase = depths
_UpperCAmelCase = num_heads
_UpperCAmelCase = window_size
_UpperCAmelCase = mlp_ratio
_UpperCAmelCase = qkv_bias
_UpperCAmelCase = hidden_dropout_prob
_UpperCAmelCase = attention_probs_dropout_prob
_UpperCAmelCase = drop_path_rate
_UpperCAmelCase = hidden_act
_UpperCAmelCase = use_absolute_embeddings
_UpperCAmelCase = patch_norm
_UpperCAmelCase = layer_norm_eps
_UpperCAmelCase = initializer_range
_UpperCAmelCase = is_training
_UpperCAmelCase = scope
_UpperCAmelCase = use_labels
_UpperCAmelCase = type_sequence_label_size
_UpperCAmelCase = encoder_stride
_UpperCAmelCase = out_features
_UpperCAmelCase = out_indices
def UpperCamelCase( self ):
_UpperCAmelCase = floats_tensor([self.batch_size, self.num_channels, self.image_size, self.image_size] )
_UpperCAmelCase = None
if self.use_labels:
_UpperCAmelCase = ids_tensor([self.batch_size] , self.type_sequence_label_size )
_UpperCAmelCase = self.get_config()
return config, pixel_values, labels
def UpperCamelCase( self ):
return FocalNetConfig(
image_size=self.image_size , patch_size=self.patch_size , num_channels=self.num_channels , embed_dim=self.embed_dim , hidden_sizes=self.hidden_sizes , depths=self.depths , num_heads=self.num_heads , window_size=self.window_size , mlp_ratio=self.mlp_ratio , qkv_bias=self.qkv_bias , hidden_dropout_prob=self.hidden_dropout_prob , attention_probs_dropout_prob=self.attention_probs_dropout_prob , drop_path_rate=self.drop_path_rate , hidden_act=self.hidden_act , use_absolute_embeddings=self.use_absolute_embeddings , path_norm=self.patch_norm , layer_norm_eps=self.layer_norm_eps , initializer_range=self.initializer_range , encoder_stride=self.encoder_stride , out_features=self.out_features , out_indices=self.out_indices , )
def UpperCamelCase( self , _UpperCamelCase , _UpperCamelCase , _UpperCamelCase ):
_UpperCAmelCase = FocalNetModel(config=_UpperCamelCase )
model.to(_UpperCamelCase )
model.eval()
_UpperCAmelCase = model(_UpperCamelCase )
_UpperCAmelCase = ((config.image_size // config.patch_size) ** 2) // (4 ** (len(config.depths ) - 1))
_UpperCAmelCase = int(config.embed_dim * 2 ** (len(config.depths ) - 1) )
self.parent.assertEqual(result.last_hidden_state.shape , (self.batch_size, expected_seq_len, expected_dim) )
def UpperCamelCase( self , _UpperCamelCase , _UpperCamelCase , _UpperCamelCase ):
_UpperCAmelCase = FocalNetBackbone(config=_UpperCamelCase )
model.to(_UpperCamelCase )
model.eval()
_UpperCAmelCase = model(_UpperCamelCase )
# verify feature maps
self.parent.assertEqual(len(result.feature_maps ) , len(config.out_features ) )
self.parent.assertListEqual(list(result.feature_maps[0].shape ) , [self.batch_size, self.image_size, 8, 8] )
# verify channels
self.parent.assertEqual(len(model.channels ) , len(config.out_features ) )
self.parent.assertListEqual(model.channels , config.hidden_sizes[:-1] )
# verify backbone works with out_features=None
_UpperCAmelCase = None
_UpperCAmelCase = FocalNetBackbone(config=_UpperCamelCase )
model.to(_UpperCamelCase )
model.eval()
_UpperCAmelCase = model(_UpperCamelCase )
# verify feature maps
self.parent.assertEqual(len(result.feature_maps ) , 1 )
self.parent.assertListEqual(list(result.feature_maps[0].shape ) , [self.batch_size, self.image_size * 2, 4, 4] )
# verify channels
self.parent.assertEqual(len(model.channels ) , 1 )
self.parent.assertListEqual(model.channels , [config.hidden_sizes[-1]] )
def UpperCamelCase( self , _UpperCamelCase , _UpperCamelCase , _UpperCamelCase ):
_UpperCAmelCase = FocalNetForMaskedImageModeling(config=_UpperCamelCase )
model.to(_UpperCamelCase )
model.eval()
_UpperCAmelCase = model(_UpperCamelCase )
self.parent.assertEqual(
result.reconstruction.shape , (self.batch_size, self.num_channels, self.image_size, self.image_size) )
# test greyscale images
_UpperCAmelCase = 1
_UpperCAmelCase = FocalNetForMaskedImageModeling(_UpperCamelCase )
model.to(_UpperCamelCase )
model.eval()
_UpperCAmelCase = floats_tensor([self.batch_size, 1, self.image_size, self.image_size] )
_UpperCAmelCase = model(_UpperCamelCase )
self.parent.assertEqual(result.reconstruction.shape , (self.batch_size, 1, self.image_size, self.image_size) )
def UpperCamelCase( self , _UpperCamelCase , _UpperCamelCase , _UpperCamelCase ):
_UpperCAmelCase = self.type_sequence_label_size
_UpperCAmelCase = FocalNetForImageClassification(_UpperCamelCase )
model.to(_UpperCamelCase )
model.eval()
_UpperCAmelCase = model(_UpperCamelCase , labels=_UpperCamelCase )
self.parent.assertEqual(result.logits.shape , (self.batch_size, self.type_sequence_label_size) )
# test greyscale images
_UpperCAmelCase = 1
_UpperCAmelCase = FocalNetForImageClassification(_UpperCamelCase )
model.to(_UpperCamelCase )
model.eval()
_UpperCAmelCase = floats_tensor([self.batch_size, 1, self.image_size, self.image_size] )
_UpperCAmelCase = model(_UpperCamelCase )
self.parent.assertEqual(result.logits.shape , (self.batch_size, self.type_sequence_label_size) )
def UpperCamelCase( self ):
_UpperCAmelCase = self.prepare_config_and_inputs()
_UpperCAmelCase , _UpperCAmelCase , _UpperCAmelCase = config_and_inputs
_UpperCAmelCase = {'''pixel_values''': pixel_values}
return config, inputs_dict
@require_torch
class __UpperCamelCase ( A__ , A__ , unittest.TestCase ):
__A : List[str] = (
(
FocalNetModel,
FocalNetForImageClassification,
FocalNetForMaskedImageModeling,
FocalNetBackbone,
)
if is_torch_available()
else ()
)
__A : Union[str, Any] = (
{"""feature-extraction""": FocalNetModel, """image-classification""": FocalNetForImageClassification}
if is_torch_available()
else {}
)
__A : int = False
__A : int = False
__A : Dict = False
__A : str = False
__A : Tuple = False
def UpperCamelCase( self ):
_UpperCAmelCase = FocalNetModelTester(self )
_UpperCAmelCase = ConfigTester(self , config_class=_UpperCamelCase , embed_dim=37 , has_text_modality=_UpperCamelCase )
def UpperCamelCase( self ):
self.create_and_test_config_common_properties()
self.config_tester.create_and_test_config_to_json_string()
self.config_tester.create_and_test_config_to_json_file()
self.config_tester.create_and_test_config_from_and_save_pretrained()
self.config_tester.create_and_test_config_with_num_labels()
self.config_tester.check_config_can_be_init_without_params()
self.config_tester.check_config_arguments_init()
def UpperCamelCase( self ):
return
def UpperCamelCase( self ):
_UpperCAmelCase = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*_UpperCamelCase )
def UpperCamelCase( self ):
_UpperCAmelCase = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_backbone(*_UpperCamelCase )
def UpperCamelCase( self ):
_UpperCAmelCase = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_masked_image_modeling(*_UpperCamelCase )
def UpperCamelCase( self ):
_UpperCAmelCase = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_image_classification(*_UpperCamelCase )
@unittest.skip(reason='''FocalNet does not use inputs_embeds''' )
def UpperCamelCase( self ):
pass
@unittest.skip(reason='''FocalNet does not use feedforward chunking''' )
def UpperCamelCase( self ):
pass
def UpperCamelCase( self ):
_UpperCAmelCase , _UpperCAmelCase = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes[:-1]:
_UpperCAmelCase = model_class(_UpperCamelCase )
self.assertIsInstance(model.get_input_embeddings() , (nn.Module) )
_UpperCAmelCase = model.get_output_embeddings()
self.assertTrue(x is None or isinstance(_UpperCamelCase , nn.Linear ) )
def UpperCamelCase( self ):
_UpperCAmelCase , _UpperCAmelCase = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes[:-1]:
_UpperCAmelCase = model_class(_UpperCamelCase )
_UpperCAmelCase = inspect.signature(model.forward )
# signature.parameters is an OrderedDict => so arg_names order is deterministic
_UpperCAmelCase = [*signature.parameters.keys()]
_UpperCAmelCase = ['''pixel_values''']
self.assertListEqual(arg_names[:1] , _UpperCamelCase )
def UpperCamelCase( self , _UpperCamelCase , _UpperCamelCase , _UpperCamelCase , _UpperCamelCase ):
_UpperCAmelCase = model_class(_UpperCamelCase )
model.to(_UpperCamelCase )
model.eval()
with torch.no_grad():
_UpperCAmelCase = model(**self._prepare_for_class(_UpperCamelCase , _UpperCamelCase ) )
_UpperCAmelCase = outputs.hidden_states
_UpperCAmelCase = getattr(
self.model_tester , '''expected_num_hidden_layers''' , len(self.model_tester.depths ) + 1 )
self.assertEqual(len(_UpperCamelCase ) , _UpperCamelCase )
# FocalNet has a different seq_length
_UpperCAmelCase = (
config.patch_size
if isinstance(config.patch_size , collections.abc.Iterable )
else (config.patch_size, config.patch_size)
)
_UpperCAmelCase = (image_size[1] // patch_size[1]) * (image_size[0] // patch_size[0])
self.assertListEqual(
list(hidden_states[0].shape[-2:] ) , [num_patches, self.model_tester.embed_dim] , )
_UpperCAmelCase = outputs.reshaped_hidden_states
self.assertEqual(len(_UpperCamelCase ) , _UpperCamelCase )
_UpperCAmelCase , _UpperCAmelCase , _UpperCAmelCase , _UpperCAmelCase = reshaped_hidden_states[0].shape
_UpperCAmelCase = (
reshaped_hidden_states[0].view(_UpperCamelCase , _UpperCamelCase , height * width ).permute(0 , 2 , 1 )
)
self.assertListEqual(
list(reshaped_hidden_states.shape[-2:] ) , [num_patches, self.model_tester.embed_dim] , )
def UpperCamelCase( self ):
_UpperCAmelCase , _UpperCAmelCase = self.model_tester.prepare_config_and_inputs_for_common()
_UpperCAmelCase = (
self.model_tester.image_size
if isinstance(self.model_tester.image_size , collections.abc.Iterable )
else (self.model_tester.image_size, self.model_tester.image_size)
)
for model_class in self.all_model_classes[:-1]:
_UpperCAmelCase = True
self.check_hidden_states_output(_UpperCamelCase , _UpperCamelCase , _UpperCamelCase , _UpperCamelCase )
# check that output_hidden_states also work using config
del inputs_dict["output_hidden_states"]
_UpperCAmelCase = True
self.check_hidden_states_output(_UpperCamelCase , _UpperCamelCase , _UpperCamelCase , _UpperCamelCase )
def UpperCamelCase( self ):
_UpperCAmelCase , _UpperCAmelCase = self.model_tester.prepare_config_and_inputs_for_common()
_UpperCAmelCase = 3
_UpperCAmelCase = (
self.model_tester.image_size
if isinstance(self.model_tester.image_size , collections.abc.Iterable )
else (self.model_tester.image_size, self.model_tester.image_size)
)
_UpperCAmelCase = (
config.patch_size
if isinstance(config.patch_size , collections.abc.Iterable )
else (config.patch_size, config.patch_size)
)
_UpperCAmelCase = image_size[0] + patch_size[0] - (image_size[0] % patch_size[0])
_UpperCAmelCase = image_size[1] + patch_size[1] - (image_size[1] % patch_size[1])
for model_class in self.all_model_classes[:-1]:
_UpperCAmelCase = True
self.check_hidden_states_output(_UpperCamelCase , _UpperCamelCase , _UpperCamelCase , (padded_height, padded_width) )
# check that output_hidden_states also work using config
del inputs_dict["output_hidden_states"]
_UpperCAmelCase = True
self.check_hidden_states_output(_UpperCamelCase , _UpperCamelCase , _UpperCamelCase , (padded_height, padded_width) )
@slow
def UpperCamelCase( self ):
for model_name in FOCALNET_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
_UpperCAmelCase = FocalNetModel.from_pretrained(_UpperCamelCase )
self.assertIsNotNone(_UpperCamelCase )
def UpperCamelCase( self ):
_UpperCAmelCase , _UpperCAmelCase = self.model_tester.prepare_config_and_inputs_for_common()
_UpperCAmelCase = _config_zero_init(_UpperCamelCase )
for model_class in self.all_model_classes:
_UpperCAmelCase = model_class(config=_UpperCamelCase )
for name, param in model.named_parameters():
if "embeddings" not in name and param.requires_grad:
self.assertIn(
((param.data.mean() * 1e9).round() / 1e9).item() , [0.0, 1.0] , msg=f'''Parameter {name} of model {model_class} seems not properly initialized''' , )
@require_vision
@require_torch
class __UpperCamelCase ( unittest.TestCase ):
@cached_property
def UpperCamelCase( self ):
# TODO update organization
return AutoImageProcessor.from_pretrained('''microsoft/focalnet-tiny''' ) if is_vision_available() else None
@slow
def UpperCamelCase( self ):
_UpperCAmelCase = FocalNetForImageClassification.from_pretrained('''microsoft/focalnet-tiny''' ).to(_UpperCamelCase )
_UpperCAmelCase = self.default_image_processor
_UpperCAmelCase = Image.open('''./tests/fixtures/tests_samples/COCO/000000039769.png''' )
_UpperCAmelCase = image_processor(images=_UpperCamelCase , return_tensors='''pt''' ).to(_UpperCamelCase )
# forward pass
with torch.no_grad():
_UpperCAmelCase = model(**_UpperCamelCase )
# verify the logits
_UpperCAmelCase = torch.Size((1, 1000) )
self.assertEqual(outputs.logits.shape , _UpperCamelCase )
_UpperCAmelCase = torch.tensor([0.2166, -0.4368, 0.2191] ).to(_UpperCamelCase )
self.assertTrue(torch.allclose(outputs.logits[0, :3] , _UpperCamelCase , atol=1e-4 ) )
self.assertTrue(outputs.logits.argmax(dim=-1 ).item() , 281 )
@require_torch
class __UpperCamelCase ( A__ , unittest.TestCase ):
__A : Optional[int] = (FocalNetBackbone,) if is_torch_available() else ()
__A : str = FocalNetConfig
__A : Any = False
def UpperCamelCase( self ):
_UpperCAmelCase = FocalNetModelTester(self ) | 32 |
def __snake_case ( __UpperCamelCase : list ,__UpperCamelCase : int = 0 ):
"""simple docstring"""
A_ = length or len(__UpperCamelCase )
A_ = False
for i in range(length - 1 ):
if list_data[i] > list_data[i + 1]:
A_ , A_ = list_data[i + 1], list_data[i]
A_ = True
return list_data if not swapped else bubble_sort(__UpperCamelCase ,length - 1 )
if __name__ == "__main__":
import doctest
doctest.testmod() | 86 | 0 |
from typing import TYPE_CHECKING
from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_tf_available, is_torch_available
lowerCamelCase__ : List[Any] = {
"""configuration_rag""": ["""RagConfig"""],
"""retrieval_rag""": ["""RagRetriever"""],
"""tokenization_rag""": ["""RagTokenizer"""],
}
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
lowerCamelCase__ : Optional[Any] = [
"""RagModel""",
"""RagPreTrainedModel""",
"""RagSequenceForGeneration""",
"""RagTokenForGeneration""",
]
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
lowerCamelCase__ : Any = [
"""TFRagModel""",
"""TFRagPreTrainedModel""",
"""TFRagSequenceForGeneration""",
"""TFRagTokenForGeneration""",
]
if TYPE_CHECKING:
from .configuration_rag import RagConfig
from .retrieval_rag import RagRetriever
from .tokenization_rag import RagTokenizer
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_rag import RagModel, RagPreTrainedModel, RagSequenceForGeneration, RagTokenForGeneration
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tf_rag import (
TFRagModel,
TFRagPreTrainedModel,
TFRagSequenceForGeneration,
TFRagTokenForGeneration,
)
else:
import sys
lowerCamelCase__ : int = _LazyModule(__name__, globals()["""__file__"""], _import_structure, module_spec=__spec__)
| 33 |
import pickle
import unittest
import torch
from accelerate import Accelerator
from accelerate.state import AcceleratorState
from accelerate.test_utils import require_cpu
@require_cpu
class _a ( unittest.TestCase ):
"""simple docstring"""
def __A ( self : List[str] ):
A_ = torch.nn.Linear(10 , 10 )
A_ = torch.optim.SGD(model.parameters() , 0.1 )
A_ = Accelerator()
A_ = accelerator.prepare(UpperCAmelCase )
try:
pickle.loads(pickle.dumps(UpperCAmelCase ) )
except Exception as e:
self.fail(f'''Accelerated optimizer pickling failed with {e}''' )
AcceleratorState._reset_state() | 86 | 0 |
"""simple docstring"""
import unittest
import numpy as np
from transformers import RobertaConfig, is_flax_available
from transformers.testing_utils import require_flax, slow
from ...test_modeling_flax_common import FlaxModelTesterMixin, floats_tensor, ids_tensor, random_attention_mask
if is_flax_available():
from transformers.models.roberta.modeling_flax_roberta import (
FlaxRobertaForCausalLM,
FlaxRobertaForMaskedLM,
FlaxRobertaForMultipleChoice,
FlaxRobertaForQuestionAnswering,
FlaxRobertaForSequenceClassification,
FlaxRobertaForTokenClassification,
FlaxRobertaModel,
)
class snake_case_ ( unittest.TestCase ):
"""simple docstring"""
def __init__( self , lowerCamelCase_ , lowerCamelCase_=1_3 , lowerCamelCase_=7 , lowerCamelCase_=True , lowerCamelCase_=True , lowerCamelCase_=True , lowerCamelCase_=True , lowerCamelCase_=9_9 , lowerCamelCase_=3_2 , lowerCamelCase_=5 , lowerCamelCase_=4 , lowerCamelCase_=3_7 , lowerCamelCase_="gelu" , lowerCamelCase_=0.1 , lowerCamelCase_=0.1 , lowerCamelCase_=5_1_2 , lowerCamelCase_=1_6 , lowerCamelCase_=2 , lowerCamelCase_=0.02 , lowerCamelCase_=4 , ) -> List[Any]:
UpperCamelCase = parent
UpperCamelCase = batch_size
UpperCamelCase = seq_length
UpperCamelCase = is_training
UpperCamelCase = use_attention_mask
UpperCamelCase = use_token_type_ids
UpperCamelCase = use_labels
UpperCamelCase = vocab_size
UpperCamelCase = hidden_size
UpperCamelCase = num_hidden_layers
UpperCamelCase = num_attention_heads
UpperCamelCase = intermediate_size
UpperCamelCase = hidden_act
UpperCamelCase = hidden_dropout_prob
UpperCamelCase = attention_probs_dropout_prob
UpperCamelCase = max_position_embeddings
UpperCamelCase = type_vocab_size
UpperCamelCase = type_sequence_label_size
UpperCamelCase = initializer_range
UpperCamelCase = num_choices
def UpperCAmelCase__ ( self) -> List[str]:
UpperCamelCase = ids_tensor([self.batch_size, self.seq_length] , self.vocab_size)
UpperCamelCase = None
if self.use_attention_mask:
UpperCamelCase = random_attention_mask([self.batch_size, self.seq_length])
UpperCamelCase = None
if self.use_token_type_ids:
UpperCamelCase = ids_tensor([self.batch_size, self.seq_length] , self.type_vocab_size)
UpperCamelCase = RobertaConfig(
vocab_size=self.vocab_size , hidden_size=self.hidden_size , num_hidden_layers=self.num_hidden_layers , num_attention_heads=self.num_attention_heads , intermediate_size=self.intermediate_size , hidden_act=self.hidden_act , hidden_dropout_prob=self.hidden_dropout_prob , attention_probs_dropout_prob=self.attention_probs_dropout_prob , max_position_embeddings=self.max_position_embeddings , type_vocab_size=self.type_vocab_size , is_decoder=lowerCamelCase_ , initializer_range=self.initializer_range , )
return config, input_ids, token_type_ids, attention_mask
def UpperCAmelCase__ ( self) -> Any:
UpperCamelCase = self.prepare_config_and_inputs()
UpperCamelCase , UpperCamelCase , UpperCamelCase , UpperCamelCase = config_and_inputs
UpperCamelCase = {'''input_ids''': input_ids, '''token_type_ids''': token_type_ids, '''attention_mask''': attention_mask}
return config, inputs_dict
def UpperCAmelCase__ ( self) -> str:
UpperCamelCase = self.prepare_config_and_inputs()
UpperCamelCase , UpperCamelCase , UpperCamelCase , UpperCamelCase = config_and_inputs
UpperCamelCase = True
UpperCamelCase = floats_tensor([self.batch_size, self.seq_length, self.hidden_size])
UpperCamelCase = ids_tensor([self.batch_size, self.seq_length] , vocab_size=2)
return (
config,
input_ids,
token_type_ids,
encoder_hidden_states,
encoder_attention_mask,
)
@require_flax
class snake_case_ ( lowerCamelCase_ , unittest.TestCase ):
"""simple docstring"""
A_ = True
A_ = (
(
FlaxRobertaModel,
FlaxRobertaForCausalLM,
FlaxRobertaForMaskedLM,
FlaxRobertaForSequenceClassification,
FlaxRobertaForTokenClassification,
FlaxRobertaForMultipleChoice,
FlaxRobertaForQuestionAnswering,
)
if is_flax_available()
else ()
)
def UpperCAmelCase__ ( self) -> Optional[int]:
UpperCamelCase = FlaxRobertaModelTester(self)
@slow
def UpperCAmelCase__ ( self) -> Dict:
for model_class_name in self.all_model_classes:
UpperCamelCase = model_class_name.from_pretrained('''roberta-base''' , from_pt=lowerCamelCase_)
UpperCamelCase = model(np.ones((1, 1)))
self.assertIsNotNone(lowerCamelCase_) | 34 |
import argparse
import json
import os
import fairseq
import torch
from torch import nn
from transformers import (
SpeechaTextaConfig,
SpeechaTextaForCausalLM,
SpeechaTextaTokenizer,
SpeechEncoderDecoderConfig,
SpeechEncoderDecoderModel,
WavaVecaConfig,
WavaVecaFeatureExtractor,
WavaVecaModel,
logging,
)
logging.set_verbosity_info()
__a :List[str] = logging.get_logger(__name__)
__a :Optional[int] = {
'post_extract_proj': 'feature_projection.projection',
'encoder.pos_conv.0': 'encoder.pos_conv_embed.conv',
'self_attn.k_proj': 'encoder.layers.*.attention.k_proj',
'self_attn.v_proj': 'encoder.layers.*.attention.v_proj',
'self_attn.q_proj': 'encoder.layers.*.attention.q_proj',
'self_attn.out_proj': 'encoder.layers.*.attention.out_proj',
'self_attn_layer_norm': 'encoder.layers.*.layer_norm',
'fc1': 'encoder.layers.*.feed_forward.intermediate_dense',
'fc2': 'encoder.layers.*.feed_forward.output_dense',
'final_layer_norm': 'encoder.layers.*.final_layer_norm',
'encoder.layer_norm': 'encoder.layer_norm',
'w2v_model.layer_norm': 'feature_projection.layer_norm',
'quantizer.weight_proj': 'quantizer.weight_proj',
'quantizer.vars': 'quantizer.codevectors',
'project_q': 'project_q',
'final_proj': 'project_hid',
'w2v_encoder.proj': 'lm_head',
'mask_emb': 'masked_spec_embed',
}
__a :Any = [
'lm_head',
'quantizer.weight_proj',
'quantizer.codevectors',
'project_q',
'project_hid',
]
def __snake_case ( __UpperCamelCase : Dict ,__UpperCamelCase : Optional[Any] ,__UpperCamelCase : Optional[int] ,__UpperCamelCase : List[str] ,__UpperCamelCase : Union[str, Any] ):
"""simple docstring"""
for attribute in key.split("." ):
A_ = getattr(__UpperCamelCase ,__UpperCamelCase )
if weight_type is not None:
A_ = getattr(__UpperCamelCase ,__UpperCamelCase ).shape
else:
A_ = hf_pointer.shape
assert hf_shape == value.shape, (
f'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
f''' {value.shape} for {full_name}'''
)
if weight_type == "weight":
A_ = value
elif weight_type == "weight_g":
A_ = value
elif weight_type == "weight_v":
A_ = value
elif weight_type == "bias":
A_ = value
else:
A_ = value
logger.info(f'''{key + "." + weight_type if weight_type is not None else ""} was initialized from {full_name}.''' )
def __snake_case ( __UpperCamelCase : Union[str, Any] ,__UpperCamelCase : Dict ):
"""simple docstring"""
A_ = []
A_ = fairseq_model.state_dict()
A_ = hf_model.feature_extractor
# if encoder has different dim to decoder -> use proj_weight
A_ = None
for name, value in fairseq_dict.items():
A_ = False
if "conv_layers" in name:
load_conv_layer(
__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,hf_model.config.feat_extract_norm == "group" ,)
A_ = True
elif name.split("." )[0] == "proj":
A_ = fairseq_model.proj
A_ = True
else:
for key, mapped_key in MAPPING.items():
if key in name or key.split("w2v_model." )[-1] == name.split("." )[0]:
A_ = True
if "*" in mapped_key:
A_ = name.split(__UpperCamelCase )[0].split("." )[-2]
A_ = mapped_key.replace("*" ,__UpperCamelCase )
if "weight_g" in name:
A_ = "weight_g"
elif "weight_v" in name:
A_ = "weight_v"
elif "bias" in name:
A_ = "bias"
elif "weight" in name:
A_ = "weight"
else:
A_ = None
set_recursively(__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase ,__UpperCamelCase )
continue
if not is_used:
unused_weights.append(__UpperCamelCase )
logger.warning(f'''Unused weights: {unused_weights}''' )
return proj_weight
def __snake_case ( __UpperCamelCase : str ,__UpperCamelCase : Optional[Any] ,__UpperCamelCase : int ,__UpperCamelCase : int ,__UpperCamelCase : Any ):
"""simple docstring"""
A_ = full_name.split("conv_layers." )[-1]
A_ = name.split("." )
A_ = int(items[0] )
A_ = int(items[1] )
if type_id == 0:
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.bias.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape, (
f'''{full_name} has size {value.shape}, but {feature_extractor[layer_id].layer_norm.bias.data.shape} was'''
" found."
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape, (
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor[layer_id].layer_norm.weight.data.shape} was found.'''
)
A_ = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(__UpperCamelCase )
def __snake_case ( __UpperCamelCase : Optional[Any] ):
"""simple docstring"""
A_ , A_ = emb.weight.shape
A_ = nn.Linear(__UpperCamelCase ,__UpperCamelCase ,bias=__UpperCamelCase )
A_ = emb.weight.data
return lin_layer
def __snake_case ( __UpperCamelCase : Tuple ):
"""simple docstring"""
with open(__UpperCamelCase ,"r" ,encoding="utf-8" ) as f:
A_ = f.readlines()
A_ = [line.split(" " )[0] for line in lines]
A_ = len(__UpperCamelCase )
A_ = {
"<s>": 0,
"<pad>": 1,
"</s>": 2,
"<unk>": 3,
}
vocab_dict.update(dict(zip(__UpperCamelCase ,range(4 ,num_words + 4 ) ) ) )
return vocab_dict
@torch.no_grad()
def __snake_case ( __UpperCamelCase : List[str] ,__UpperCamelCase : Any ,__UpperCamelCase : List[Any] ,__UpperCamelCase : Union[str, Any] ,__UpperCamelCase : List[str] ,__UpperCamelCase : Optional[Any] ,__UpperCamelCase : Dict ,):
"""simple docstring"""
A_ = WavaVecaConfig.from_pretrained(__UpperCamelCase )
A_ = SpeechaTextaConfig.from_pretrained(
__UpperCamelCase ,vocab_size=__UpperCamelCase ,decoder_layers=__UpperCamelCase ,do_stable_layer_norm=__UpperCamelCase )
A_ = WavaVecaFeatureExtractor(
feature_size=1 ,sampling_rate=1_6000 ,padding_value=0 ,do_normalize=__UpperCamelCase ,return_attention_mask=__UpperCamelCase ,)
A_ , A_ , A_ = fairseq.checkpoint_utils.load_model_ensemble_and_task(
[checkpoint_path] ,arg_overrides={"data": "/".join(dict_path.split("/" )[:-1] )} )
A_ = model[0].eval()
# set weights for wav2vec2 encoder
A_ = WavaVecaModel(__UpperCamelCase )
A_ = recursively_load_weights_wavaveca(model.encoder ,__UpperCamelCase )
A_ = SpeechaTextaForCausalLM(__UpperCamelCase )
A_ , A_ = hf_decoder.model.decoder.load_state_dict(model.decoder.state_dict() ,strict=__UpperCamelCase )
# set output linear layer
unexpected_keys.remove("embed_out" )
A_ = nn.Parameter(model.decoder.embed_out.detach() )
# layer norm is init to identity matrix so leaving it is fine
logger.warning(f'''The following keys are missing when loading the decoder weights: {missing_keys}''' )
logger.warning(f'''The following keys are unexpected when loading the decoder weights: {unexpected_keys}''' )
A_ = SpeechEncoderDecoderModel(encoder=__UpperCamelCase ,decoder=__UpperCamelCase )
A_ = False
# add projection layer
A_ = nn.Parameter(projection_layer.weight )
A_ = nn.Parameter(projection_layer.bias )
A_ = create_vocab_dict(__UpperCamelCase )
with open(os.path.join(__UpperCamelCase ,"vocab.json" ) ,"w" ) as fp:
json.dump(__UpperCamelCase ,__UpperCamelCase )
A_ = SpeechaTextaTokenizer(os.path.join(__UpperCamelCase ,"vocab.json" ) )
tokenizer.save_pretrained(__UpperCamelCase )
A_ = hf_wavavec.config.to_dict()
A_ = tokenizer.pad_token_id
A_ = tokenizer.bos_token_id
A_ = tokenizer.eos_token_id
A_ = "speech_to_text_2"
A_ = "wav2vec2"
A_ = SpeechEncoderDecoderConfig.from_dict(__UpperCamelCase )
hf_wavavec.save_pretrained(__UpperCamelCase )
feature_extractor.save_pretrained(__UpperCamelCase )
if __name__ == "__main__":
__a :int = argparse.ArgumentParser()
parser.add_argument('--pytorch_dump_folder_path', default=None, type=str, help='Path to the output PyTorch model.')
parser.add_argument('--checkpoint_path', default=None, type=str, help='Path to fairseq checkpoint')
parser.add_argument('--dict_path', default=None, type=str, help='Path to dict of fine-tuned model')
parser.add_argument(
'--encoder_config_path',
default='facebook/wav2vec2-large-lv60',
type=str,
help='Path to hf encoder wav2vec2 checkpoint config',
)
parser.add_argument(
'--decoder_config_path',
default='facebook/s2t-small-mustc-en-fr-st',
type=str,
help='Path to hf decoder s2t checkpoint config',
)
parser.add_argument('--vocab_size', default=1_0224, type=int, help='Vocab size of decoder')
parser.add_argument('--num_decoder_layers', default=7, type=int, help='Number of decoder layers')
__a :Tuple = parser.parse_args()
convert_wavaveca_checkpoint(
args.checkpoint_path,
args.pytorch_dump_folder_path,
args.dict_path,
encoder_config_path=args.encoder_config_path,
decoder_config_path=args.decoder_config_path,
vocab_size=args.vocab_size,
num_decoder_layers=args.num_decoder_layers,
) | 86 | 0 |
from typing import TYPE_CHECKING
from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_tokenizers_available, is_torch_available
a_ :Optional[Any] = {
'configuration_roc_bert': ['ROC_BERT_PRETRAINED_CONFIG_ARCHIVE_MAP', 'RoCBertConfig'],
'tokenization_roc_bert': ['RoCBertTokenizer'],
}
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
pass
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :Union[str, Any] = [
'ROC_BERT_PRETRAINED_MODEL_ARCHIVE_LIST',
'RoCBertForCausalLM',
'RoCBertForMaskedLM',
'RoCBertForMultipleChoice',
'RoCBertForPreTraining',
'RoCBertForQuestionAnswering',
'RoCBertForSequenceClassification',
'RoCBertForTokenClassification',
'RoCBertLayer',
'RoCBertModel',
'RoCBertPreTrainedModel',
'load_tf_weights_in_roc_bert',
]
if TYPE_CHECKING:
from .configuration_roc_bert import ROC_BERT_PRETRAINED_CONFIG_ARCHIVE_MAP, RoCBertConfig
from .tokenization_roc_bert import RoCBertTokenizer
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
raise OptionalDependencyNotAvailable()
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_roc_bert import (
ROC_BERT_PRETRAINED_MODEL_ARCHIVE_LIST,
RoCBertForCausalLM,
RoCBertForMaskedLM,
RoCBertForMultipleChoice,
RoCBertForPreTraining,
RoCBertForQuestionAnswering,
RoCBertForSequenceClassification,
RoCBertForTokenClassification,
RoCBertLayer,
RoCBertModel,
RoCBertPreTrainedModel,
load_tf_weights_in_roc_bert,
)
else:
import sys
a_ :List[Any] = _LazyModule(__name__, globals()['__file__'], _import_structure, module_spec=__spec__)
| 35 |
from typing import Any, Dict, List, Union
from ..utils import add_end_docstrings, is_torch_available, is_vision_available, logging, requires_backends
from .base import PIPELINE_INIT_ARGS, Pipeline
if is_vision_available():
from ..image_utils import load_image
if is_torch_available():
import torch
from ..models.auto.modeling_auto import MODEL_FOR_OBJECT_DETECTION_MAPPING, MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING
__a :str = logging.get_logger(__name__)
__a :Any = Dict[str, Any]
__a :int = List[Prediction]
@add_end_docstrings(snake_case_ )
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Tuple , *UpperCAmelCase : Optional[int] , **UpperCAmelCase : Optional[Any] ):
super().__init__(*UpperCAmelCase , **UpperCAmelCase )
if self.framework == "tf":
raise ValueError(f'''The {self.__class__} is only available in PyTorch.''' )
requires_backends(self , "vision" )
self.check_model_type(
dict(MODEL_FOR_OBJECT_DETECTION_MAPPING.items() + MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING.items() ) )
def __A ( self : str , **UpperCAmelCase : str ):
A_ = {}
if "threshold" in kwargs:
A_ = kwargs["threshold"]
return {}, {}, postprocess_kwargs
def __call__( self : Union[str, Any] , *UpperCAmelCase : Union[str, Any] , **UpperCAmelCase : Optional[Any] ):
return super().__call__(*UpperCAmelCase , **UpperCAmelCase )
def __A ( self : str , UpperCAmelCase : Any ):
A_ = load_image(UpperCAmelCase )
A_ = torch.IntTensor([[image.height, image.width]] )
A_ = self.image_processor(images=[image] , return_tensors="pt" )
if self.tokenizer is not None:
A_ = self.tokenizer(text=inputs["words"] , boxes=inputs["boxes"] , return_tensors="pt" )
A_ = target_size
return inputs
def __A ( self : Optional[Any] , UpperCAmelCase : Optional[int] ):
A_ = model_inputs.pop("target_size" )
A_ = self.model(**UpperCAmelCase )
A_ = outputs.__class__({"target_size": target_size, **outputs} )
if self.tokenizer is not None:
A_ = model_inputs["bbox"]
return model_outputs
def __A ( self : str , UpperCAmelCase : Optional[int] , UpperCAmelCase : Union[str, Any]=0.9 ):
A_ = model_outputs["target_size"]
if self.tokenizer is not None:
# This is a LayoutLMForTokenClassification variant.
# The OCR got the boxes and the model classified the words.
A_ , A_ = target_size[0].tolist()
def unnormalize(UpperCAmelCase : Any ):
return self._get_bounding_box(
torch.Tensor(
[
(width * bbox[0] / 1000),
(height * bbox[1] / 1000),
(width * bbox[2] / 1000),
(height * bbox[3] / 1000),
] ) )
A_ , A_ = model_outputs["logits"].squeeze(0 ).softmax(dim=-1 ).max(dim=-1 )
A_ = [self.model.config.idalabel[prediction] for prediction in classes.tolist()]
A_ = [unnormalize(UpperCAmelCase ) for bbox in model_outputs["bbox"].squeeze(0 )]
A_ = ["score", "label", "box"]
A_ = [dict(zip(UpperCAmelCase , UpperCAmelCase ) ) for vals in zip(scores.tolist() , UpperCAmelCase , UpperCAmelCase ) if vals[0] > threshold]
else:
# This is a regular ForObjectDetectionModel
A_ = self.image_processor.post_process_object_detection(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase )
A_ = raw_annotations[0]
A_ = raw_annotation["scores"]
A_ = raw_annotation["labels"]
A_ = raw_annotation["boxes"]
A_ = scores.tolist()
A_ = [self.model.config.idalabel[label.item()] for label in labels]
A_ = [self._get_bounding_box(UpperCAmelCase ) for box in boxes]
# {"scores": [...], ...} --> [{"score":x, ...}, ...]
A_ = ["score", "label", "box"]
A_ = [
dict(zip(UpperCAmelCase , UpperCAmelCase ) )
for vals in zip(raw_annotation["scores"] , raw_annotation["labels"] , raw_annotation["boxes"] )
]
return annotation
def __A ( self : Tuple , UpperCAmelCase : "torch.Tensor" ):
if self.framework != "pt":
raise ValueError("The ObjectDetectionPipeline is only available in PyTorch." )
A_ , A_ , A_ , A_ = box.int().tolist()
A_ = {
"xmin": xmin,
"ymin": ymin,
"xmax": xmax,
"ymax": ymax,
}
return bbox | 86 | 0 |
import argparse
from diffusers.pipelines.stable_diffusion.convert_from_ckpt import download_controlnet_from_original_ckpt
if __name__ == "__main__":
__lowercase : Dict = argparse.ArgumentParser()
parser.add_argument(
'''--checkpoint_path''', default=None, type=str, required=True, help='''Path to the checkpoint to convert.'''
)
parser.add_argument(
'''--original_config_file''',
type=str,
required=True,
help='''The YAML config file corresponding to the original architecture.''',
)
parser.add_argument(
'''--num_in_channels''',
default=None,
type=int,
help='''The number of input channels. If `None` number of input channels will be automatically inferred.''',
)
parser.add_argument(
'''--image_size''',
default=512,
type=int,
help=(
'''The image size that the model was trained on. Use 512 for Stable Diffusion v1.X and Stable Siffusion v2'''
''' Base. Use 768 for Stable Diffusion v2.'''
),
)
parser.add_argument(
'''--extract_ema''',
action='''store_true''',
help=(
'''Only relevant for checkpoints that have both EMA and non-EMA weights. Whether to extract the EMA weights'''
''' or not. Defaults to `False`. Add `--extract_ema` to extract the EMA weights. EMA weights usually yield'''
''' higher quality images for inference. Non-EMA weights are usually better to continue fine-tuning.'''
),
)
parser.add_argument(
'''--upcast_attention''',
action='''store_true''',
help=(
'''Whether the attention computation should always be upcasted. This is necessary when running stable'''
''' diffusion 2.1.'''
),
)
parser.add_argument(
'''--from_safetensors''',
action='''store_true''',
help='''If `--checkpoint_path` is in `safetensors` format, load checkpoint with safetensors instead of PyTorch.''',
)
parser.add_argument(
'''--to_safetensors''',
action='''store_true''',
help='''Whether to store pipeline in safetensors format or not.''',
)
parser.add_argument('''--dump_path''', default=None, type=str, required=True, help='''Path to the output model.''')
parser.add_argument('''--device''', type=str, help='''Device to use (e.g. cpu, cuda:0, cuda:1, etc.)''')
def lowercase ( __A : Tuple ) -> Union[str, Any]:
'''simple docstring'''
if string == "True":
return True
elif string == "False":
return False
else:
raise ValueError(f"""could not parse string as bool {string}""" )
parser.add_argument(
'''--use_linear_projection''', help='''Override for use linear projection''', required=False, type=parse_bool
)
parser.add_argument('''--cross_attention_dim''', help='''Override for cross attention_dim''', required=False, type=int)
__lowercase : Union[str, Any] = parser.parse_args()
__lowercase : Optional[int] = download_controlnet_from_original_ckpt(
checkpoint_path=args.checkpoint_path,
original_config_file=args.original_config_file,
image_size=args.image_size,
extract_ema=args.extract_ema,
num_in_channels=args.num_in_channels,
upcast_attention=args.upcast_attention,
from_safetensors=args.from_safetensors,
device=args.device,
use_linear_projection=args.use_linear_projection,
cross_attention_dim=args.cross_attention_dim,
)
controlnet.save_pretrained(args.dump_path, safe_serialization=args.to_safetensors)
| 36 |
import inspect
from typing import List, Optional, Tuple, Union
import numpy as np
import PIL
import torch
import torch.utils.checkpoint
from ...models import UNetaDModel, VQModel
from ...schedulers import (
DDIMScheduler,
DPMSolverMultistepScheduler,
EulerAncestralDiscreteScheduler,
EulerDiscreteScheduler,
LMSDiscreteScheduler,
PNDMScheduler,
)
from ...utils import PIL_INTERPOLATION, randn_tensor
from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
def __snake_case ( __UpperCamelCase : Dict ):
"""simple docstring"""
A_ , A_ = image.size
A_ , A_ = (x - x % 32 for x in (w, h)) # resize to integer multiple of 32
A_ = image.resize((w, h) ,resample=PIL_INTERPOLATION["lanczos"] )
A_ = np.array(__UpperCamelCase ).astype(np.floataa ) / 255.0
A_ = image[None].transpose(0 ,3 ,1 ,2 )
A_ = torch.from_numpy(__UpperCamelCase )
return 2.0 * image - 1.0
class _a ( snake_case_ ):
"""simple docstring"""
def __init__( self : Union[str, Any] , UpperCAmelCase : VQModel , UpperCAmelCase : UNetaDModel , UpperCAmelCase : Union[
DDIMScheduler,
PNDMScheduler,
LMSDiscreteScheduler,
EulerDiscreteScheduler,
EulerAncestralDiscreteScheduler,
DPMSolverMultistepScheduler,
] , ):
super().__init__()
self.register_modules(vqvae=UpperCAmelCase , unet=UpperCAmelCase , scheduler=UpperCAmelCase )
@torch.no_grad()
def __call__( self : int , UpperCAmelCase : Union[torch.Tensor, PIL.Image.Image] = None , UpperCAmelCase : Optional[int] = 1 , UpperCAmelCase : Optional[int] = 100 , UpperCAmelCase : Optional[float] = 0.0 , UpperCAmelCase : Optional[Union[torch.Generator, List[torch.Generator]]] = None , UpperCAmelCase : Optional[str] = "pil" , UpperCAmelCase : bool = True , ):
if isinstance(UpperCAmelCase , PIL.Image.Image ):
A_ = 1
elif isinstance(UpperCAmelCase , torch.Tensor ):
A_ = image.shape[0]
else:
raise ValueError(f'''`image` has to be of type `PIL.Image.Image` or `torch.Tensor` but is {type(UpperCAmelCase )}''' )
if isinstance(UpperCAmelCase , PIL.Image.Image ):
A_ = preprocess(UpperCAmelCase )
A_ , A_ = image.shape[-2:]
# in_channels should be 6: 3 for latents, 3 for low resolution image
A_ = (batch_size, self.unet.config.in_channels // 2, height, width)
A_ = next(self.unet.parameters() ).dtype
A_ = randn_tensor(UpperCAmelCase , generator=UpperCAmelCase , device=self.device , dtype=UpperCAmelCase )
A_ = image.to(device=self.device , dtype=UpperCAmelCase )
# set timesteps and move to the correct device
self.scheduler.set_timesteps(UpperCAmelCase , device=self.device )
A_ = self.scheduler.timesteps
# scale the initial noise by the standard deviation required by the scheduler
A_ = latents * self.scheduler.init_noise_sigma
# prepare extra kwargs for the scheduler step, since not all schedulers have the same signature.
# eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
# eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
# and should be between [0, 1]
A_ = "eta" in set(inspect.signature(self.scheduler.step ).parameters.keys() )
A_ = {}
if accepts_eta:
A_ = eta
for t in self.progress_bar(UpperCAmelCase ):
# concat latents and low resolution image in the channel dimension.
A_ = torch.cat([latents, image] , dim=1 )
A_ = self.scheduler.scale_model_input(UpperCAmelCase , UpperCAmelCase )
# predict the noise residual
A_ = self.unet(UpperCAmelCase , UpperCAmelCase ).sample
# compute the previous noisy sample x_t -> x_t-1
A_ = self.scheduler.step(UpperCAmelCase , UpperCAmelCase , UpperCAmelCase , **UpperCAmelCase ).prev_sample
# decode the image latents with the VQVAE
A_ = self.vqvae.decode(UpperCAmelCase ).sample
A_ = torch.clamp(UpperCAmelCase , -1.0 , 1.0 )
A_ = image / 2 + 0.5
A_ = image.cpu().permute(0 , 2 , 3 , 1 ).numpy()
if output_type == "pil":
A_ = self.numpy_to_pil(UpperCAmelCase )
if not return_dict:
return (image,)
return ImagePipelineOutput(images=UpperCAmelCase ) | 86 | 0 |
import argparse
import os
import numpy as np
import tensorflow as tf
import torch
from transformers import BertModel
def UpperCamelCase_ ( __a , __a , __a ) -> Optional[Any]:
a__ : Union[str, Any] = ("dense.weight", "attention.self.query", "attention.self.key", "attention.self.value")
a__ : Union[str, Any] = (
("layer.", "layer_"),
("word_embeddings.weight", "word_embeddings"),
("position_embeddings.weight", "position_embeddings"),
("token_type_embeddings.weight", "token_type_embeddings"),
(".", "/"),
("LayerNorm/weight", "LayerNorm/gamma"),
("LayerNorm/bias", "LayerNorm/beta"),
("weight", "kernel"),
)
if not os.path.isdir(__a ):
os.makedirs(__a )
a__ : Any = model.state_dict()
def to_tf_var_name(__a ):
for patt, repl in iter(__a ):
a__ : Tuple = name.replace(__a , __a )
return f'''bert/{name}'''
def create_tf_var(__a , __a , __a ):
a__ : Tuple = tf.dtypes.as_dtype(tensor.dtype )
a__ : Dict = tf.get_variable(dtype=__a , shape=tensor.shape , name=__a , initializer=tf.zeros_initializer() )
session.run(tf.variables_initializer([tf_var] ) )
session.run(__a )
return tf_var
tf.reset_default_graph()
with tf.Session() as session:
for var_name in state_dict:
a__ : int = to_tf_var_name(__a )
a__ : Union[str, Any] = state_dict[var_name].numpy()
if any(x in var_name for x in tensors_to_transpose ):
a__ : int = torch_tensor.T
a__ : Optional[Any] = create_tf_var(tensor=__a , name=__a , session=__a )
tf.keras.backend.set_value(__a , __a )
a__ : int = session.run(__a )
print(f'''Successfully created {tf_name}: {np.allclose(__a , __a )}''' )
a__ : Any = tf.train.Saver(tf.trainable_variables() )
saver.save(__a , os.path.join(__a , model_name.replace("-" , "_" ) + ".ckpt" ) )
def UpperCamelCase_ ( __a=None ) -> int:
a__ : Dict = argparse.ArgumentParser()
parser.add_argument("--model_name" , type=__a , required=__a , help="model name e.g. bert-base-uncased" )
parser.add_argument(
"--cache_dir" , type=__a , default=__a , required=__a , help="Directory containing pytorch model" )
parser.add_argument("--pytorch_model_path" , type=__a , required=__a , help="/path/to/<pytorch-model-name>.bin" )
parser.add_argument("--tf_cache_dir" , type=__a , required=__a , help="Directory in which to save tensorflow model" )
a__ : Optional[Any] = parser.parse_args(__a )
a__ : Tuple = BertModel.from_pretrained(
pretrained_model_name_or_path=args.model_name , state_dict=torch.load(args.pytorch_model_path ) , cache_dir=args.cache_dir , )
convert_pytorch_checkpoint_to_tf(model=__a , ckpt_dir=args.tf_cache_dir , model_name=args.model_name )
if __name__ == "__main__":
main()
| 37 |
__a :Optional[int] = [sum(int(c, 10) ** 2 for c in i.__str__()) for i in range(10_0000)]
def __snake_case ( __UpperCamelCase : int ):
"""simple docstring"""
A_ = 0
while number:
# Increased Speed Slightly by checking every 5 digits together.
sum_of_digits_squared += DIGITS_SQUARED[number % 10_0000]
number //= 10_0000
return sum_of_digits_squared
# There are 2 Chains made,
# One ends with 89 with the chain member 58 being the one which when declared first,
# there will be the least number of iterations for all the members to be checked.
# The other one ends with 1 and has only one element 1.
# So 58 and 1 are chosen to be declared at the starting.
# Changed dictionary to an array to quicken the solution
__a :list[bool | None] = [None] * 1000_0000
__a :Optional[Any] = True
__a :List[Any] = False
def __snake_case ( __UpperCamelCase : int ):
"""simple docstring"""
if CHAINS[number - 1] is not None:
return CHAINS[number - 1] # type: ignore
A_ = chain(next_number(__UpperCamelCase ) )
A_ = number_chain
while number < 1000_0000:
A_ = number_chain
number *= 10
return number_chain
def __snake_case ( __UpperCamelCase : int = 1000_0000 ):
"""simple docstring"""
for i in range(1 ,__UpperCamelCase ):
if CHAINS[i] is None:
chain(i + 1 )
return CHAINS[:number].count(__UpperCamelCase )
if __name__ == "__main__":
import doctest
doctest.testmod()
print(F"{solution() = }") | 86 | 0 |
'''simple docstring'''
import gc
import random
import unittest
import numpy as np
import torch
from transformers import CLIPTextConfig, CLIPTextModel, CLIPTokenizer
from diffusers import AutoencoderKL, CycleDiffusionPipeline, DDIMScheduler, UNetaDConditionModel
from diffusers.utils import floats_tensor, load_image, load_numpy, slow, torch_device
from diffusers.utils.testing_utils import enable_full_determinism, require_torch_gpu, skip_mps
from ..pipeline_params import (
IMAGE_TO_IMAGE_IMAGE_PARAMS,
TEXT_GUIDED_IMAGE_VARIATION_BATCH_PARAMS,
TEXT_GUIDED_IMAGE_VARIATION_PARAMS,
)
from ..test_pipelines_common import PipelineLatentTesterMixin, PipelineTesterMixin
enable_full_determinism()
class __snake_case ( __SCREAMING_SNAKE_CASE , __SCREAMING_SNAKE_CASE , unittest.TestCase ):
'''simple docstring'''
lowerCamelCase__ = CycleDiffusionPipeline
lowerCamelCase__ = TEXT_GUIDED_IMAGE_VARIATION_PARAMS - {
'''negative_prompt''',
'''height''',
'''width''',
'''negative_prompt_embeds''',
}
lowerCamelCase__ = PipelineTesterMixin.required_optional_params - {'''latents'''}
lowerCamelCase__ = TEXT_GUIDED_IMAGE_VARIATION_BATCH_PARAMS.union({'''source_prompt'''} )
lowerCamelCase__ = IMAGE_TO_IMAGE_IMAGE_PARAMS
lowerCamelCase__ = IMAGE_TO_IMAGE_IMAGE_PARAMS
def __UpperCamelCase ( self ):
torch.manual_seed(0 )
snake_case__ : Tuple = UNetaDConditionModel(
block_out_channels=(3_2, 6_4) , layers_per_block=2 , sample_size=3_2 , in_channels=4 , out_channels=4 , down_block_types=("""DownBlock2D""", """CrossAttnDownBlock2D""") , up_block_types=("""CrossAttnUpBlock2D""", """UpBlock2D""") , cross_attention_dim=3_2 , )
snake_case__ : List[str] = DDIMScheduler(
beta_start=0.0_0085 , beta_end=0.012 , beta_schedule="""scaled_linear""" , num_train_timesteps=1_0_0_0 , clip_sample=__SCREAMING_SNAKE_CASE , set_alpha_to_one=__SCREAMING_SNAKE_CASE , )
torch.manual_seed(0 )
snake_case__ : List[Any] = AutoencoderKL(
block_out_channels=[3_2, 6_4] , in_channels=3 , out_channels=3 , down_block_types=["""DownEncoderBlock2D""", """DownEncoderBlock2D"""] , up_block_types=["""UpDecoderBlock2D""", """UpDecoderBlock2D"""] , latent_channels=4 , )
torch.manual_seed(0 )
snake_case__ : Any = CLIPTextConfig(
bos_token_id=0 , eos_token_id=2 , hidden_size=3_2 , intermediate_size=3_7 , layer_norm_eps=1e-0_5 , num_attention_heads=4 , num_hidden_layers=5 , pad_token_id=1 , vocab_size=1_0_0_0 , )
snake_case__ : Any = CLIPTextModel(__SCREAMING_SNAKE_CASE )
snake_case__ : Tuple = CLIPTokenizer.from_pretrained("""hf-internal-testing/tiny-random-clip""" )
snake_case__ : List[Any] = {
"""unet""": unet,
"""scheduler""": scheduler,
"""vae""": vae,
"""text_encoder""": text_encoder,
"""tokenizer""": tokenizer,
"""safety_checker""": None,
"""feature_extractor""": None,
}
return components
def __UpperCamelCase ( self , __SCREAMING_SNAKE_CASE , __SCREAMING_SNAKE_CASE=0 ):
snake_case__ : List[str] = floats_tensor((1, 3, 3_2, 3_2) , rng=random.Random(__SCREAMING_SNAKE_CASE ) ).to(__SCREAMING_SNAKE_CASE )
snake_case__ : List[str] = image / 2 + 0.5
if str(__SCREAMING_SNAKE_CASE ).startswith("""mps""" ):
snake_case__ : Union[str, Any] = torch.manual_seed(__SCREAMING_SNAKE_CASE )
else:
snake_case__ : int = torch.Generator(device=__SCREAMING_SNAKE_CASE ).manual_seed(__SCREAMING_SNAKE_CASE )
snake_case__ : Any = {
"""prompt""": """An astronaut riding an elephant""",
"""source_prompt""": """An astronaut riding a horse""",
"""image""": image,
"""generator""": generator,
"""num_inference_steps""": 2,
"""eta""": 0.1,
"""strength""": 0.8,
"""guidance_scale""": 3,
"""source_guidance_scale""": 1,
"""output_type""": """numpy""",
}
return inputs
def __UpperCamelCase ( self ):
snake_case__ : Tuple = """cpu""" # ensure determinism for the device-dependent torch.Generator
snake_case__ : Union[str, Any] = self.get_dummy_components()
snake_case__ : int = CycleDiffusionPipeline(**__SCREAMING_SNAKE_CASE )
snake_case__ : Optional[Any] = pipe.to(__SCREAMING_SNAKE_CASE )
pipe.set_progress_bar_config(disable=__SCREAMING_SNAKE_CASE )
snake_case__ : List[str] = self.get_dummy_inputs(__SCREAMING_SNAKE_CASE )
snake_case__ : Any = pipe(**__SCREAMING_SNAKE_CASE )
snake_case__ : List[Any] = output.images
snake_case__ : Dict = images[0, -3:, -3:, -1]
assert images.shape == (1, 3_2, 3_2, 3)
snake_case__ : Optional[int] = np.array([0.4459, 0.4943, 0.4544, 0.6643, 0.5474, 0.4327, 0.5701, 0.5959, 0.5179] )
assert np.abs(image_slice.flatten() - expected_slice ).max() < 1e-2
@unittest.skipIf(torch_device != """cuda""" , """This test requires a GPU""" )
def __UpperCamelCase ( self ):
snake_case__ : str = self.get_dummy_components()
for name, module in components.items():
if hasattr(__SCREAMING_SNAKE_CASE , """half""" ):
snake_case__ : Dict = module.half()
snake_case__ : Any = CycleDiffusionPipeline(**__SCREAMING_SNAKE_CASE )
snake_case__ : List[Any] = pipe.to(__SCREAMING_SNAKE_CASE )
pipe.set_progress_bar_config(disable=__SCREAMING_SNAKE_CASE )
snake_case__ : Tuple = self.get_dummy_inputs(__SCREAMING_SNAKE_CASE )
snake_case__ : Optional[int] = pipe(**__SCREAMING_SNAKE_CASE )
snake_case__ : int = output.images
snake_case__ : Dict = images[0, -3:, -3:, -1]
assert images.shape == (1, 3_2, 3_2, 3)
snake_case__ : Tuple = np.array([0.3506, 0.4543, 0.446, 0.4575, 0.5195, 0.4155, 0.5273, 0.518, 0.4116] )
assert np.abs(image_slice.flatten() - expected_slice ).max() < 1e-2
@skip_mps
def __UpperCamelCase ( self ):
return super().test_save_load_local()
@unittest.skip("""non-deterministic pipeline""" )
def __UpperCamelCase ( self ):
return super().test_inference_batch_single_identical()
@skip_mps
def __UpperCamelCase ( self ):
return super().test_dict_tuple_outputs_equivalent()
@skip_mps
def __UpperCamelCase ( self ):
return super().test_save_load_optional_components()
@skip_mps
def __UpperCamelCase ( self ):
return super().test_attention_slicing_forward_pass()
@slow
@require_torch_gpu
class __snake_case ( unittest.TestCase ):
'''simple docstring'''
def __UpperCamelCase ( self ):
# clean up the VRAM after each test
super().tearDown()
gc.collect()
torch.cuda.empty_cache()
def __UpperCamelCase ( self ):
snake_case__ : Tuple = load_image(
"""https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main"""
"""/cycle-diffusion/black_colored_car.png""" )
snake_case__ : List[str] = load_numpy(
"""https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/cycle-diffusion/blue_colored_car_fp16.npy""" )
snake_case__ : Dict = init_image.resize((5_1_2, 5_1_2) )
snake_case__ : Tuple = """CompVis/stable-diffusion-v1-4"""
snake_case__ : Any = DDIMScheduler.from_pretrained(__SCREAMING_SNAKE_CASE , subfolder="""scheduler""" )
snake_case__ : Optional[Any] = CycleDiffusionPipeline.from_pretrained(
__SCREAMING_SNAKE_CASE , scheduler=__SCREAMING_SNAKE_CASE , safety_checker=__SCREAMING_SNAKE_CASE , torch_dtype=torch.floataa , revision="""fp16""" )
pipe.to(__SCREAMING_SNAKE_CASE )
pipe.set_progress_bar_config(disable=__SCREAMING_SNAKE_CASE )
pipe.enable_attention_slicing()
snake_case__ : Optional[int] = """A black colored car"""
snake_case__ : int = """A blue colored car"""
snake_case__ : Optional[Any] = torch.manual_seed(0 )
snake_case__ : Any = pipe(
prompt=__SCREAMING_SNAKE_CASE , source_prompt=__SCREAMING_SNAKE_CASE , image=__SCREAMING_SNAKE_CASE , num_inference_steps=1_0_0 , eta=0.1 , strength=0.85 , guidance_scale=3 , source_guidance_scale=1 , generator=__SCREAMING_SNAKE_CASE , output_type="""np""" , )
snake_case__ : List[Any] = output.images
# the values aren't exactly equal, but the images look the same visually
assert np.abs(image - expected_image ).max() < 5e-1
def __UpperCamelCase ( self ):
snake_case__ : Optional[int] = load_image(
"""https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main"""
"""/cycle-diffusion/black_colored_car.png""" )
snake_case__ : Dict = load_numpy(
"""https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/cycle-diffusion/blue_colored_car.npy""" )
snake_case__ : Any = init_image.resize((5_1_2, 5_1_2) )
snake_case__ : List[str] = """CompVis/stable-diffusion-v1-4"""
snake_case__ : Dict = DDIMScheduler.from_pretrained(__SCREAMING_SNAKE_CASE , subfolder="""scheduler""" )
snake_case__ : Dict = CycleDiffusionPipeline.from_pretrained(__SCREAMING_SNAKE_CASE , scheduler=__SCREAMING_SNAKE_CASE , safety_checker=__SCREAMING_SNAKE_CASE )
pipe.to(__SCREAMING_SNAKE_CASE )
pipe.set_progress_bar_config(disable=__SCREAMING_SNAKE_CASE )
pipe.enable_attention_slicing()
snake_case__ : Tuple = """A black colored car"""
snake_case__ : List[str] = """A blue colored car"""
snake_case__ : Tuple = torch.manual_seed(0 )
snake_case__ : List[str] = pipe(
prompt=__SCREAMING_SNAKE_CASE , source_prompt=__SCREAMING_SNAKE_CASE , image=__SCREAMING_SNAKE_CASE , num_inference_steps=1_0_0 , eta=0.1 , strength=0.85 , guidance_scale=3 , source_guidance_scale=1 , generator=__SCREAMING_SNAKE_CASE , output_type="""np""" , )
snake_case__ : int = output.images
assert np.abs(image - expected_image ).max() < 2e-2
| 38 |
from typing import TYPE_CHECKING
from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_tf_available, is_torch_available
__a :List[Any] = {
'configuration_tapas': ['TAPAS_PRETRAINED_CONFIG_ARCHIVE_MAP', 'TapasConfig'],
'tokenization_tapas': ['TapasTokenizer'],
}
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Any = [
'TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST',
'TapasForMaskedLM',
'TapasForQuestionAnswering',
'TapasForSequenceClassification',
'TapasModel',
'TapasPreTrainedModel',
'load_tf_weights_in_tapas',
]
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a :Dict = [
'TF_TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST',
'TFTapasForMaskedLM',
'TFTapasForQuestionAnswering',
'TFTapasForSequenceClassification',
'TFTapasModel',
'TFTapasPreTrainedModel',
]
if TYPE_CHECKING:
from .configuration_tapas import TAPAS_PRETRAINED_CONFIG_ARCHIVE_MAP, TapasConfig
from .tokenization_tapas import TapasTokenizer
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tapas import (
TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST,
TapasForMaskedLM,
TapasForQuestionAnswering,
TapasForSequenceClassification,
TapasModel,
TapasPreTrainedModel,
load_tf_weights_in_tapas,
)
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tf_tapas import (
TF_TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST,
TFTapasForMaskedLM,
TFTapasForQuestionAnswering,
TFTapasForSequenceClassification,
TFTapasModel,
TFTapasPreTrainedModel,
)
else:
import sys
__a :str = _LazyModule(__name__, globals()['__file__'], _import_structure, module_spec=__spec__) | 86 | 0 |
import argparse
import json
import requests
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from torchvision import transforms
from transformers import BitImageProcessor, FocalNetConfig, FocalNetForImageClassification
from transformers.image_utils import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD, PILImageResampling
def __SCREAMING_SNAKE_CASE (SCREAMING_SNAKE_CASE__ ):
snake_case_ = [2, 2, 6, 2] if '''tiny''' in model_name else [2, 2, 18, 2]
snake_case_ = True if '''large''' in model_name or '''huge''' in model_name else False
snake_case_ = True if '''large''' in model_name or '''huge''' in model_name else False
snake_case_ = True if '''large''' in model_name or '''huge''' in model_name else False
if "large" in model_name or "xlarge" in model_name or "huge" in model_name:
if "fl3" in model_name:
snake_case_ = [3, 3, 3, 3]
snake_case_ = [5, 5, 5, 5]
elif "fl4" in model_name:
snake_case_ = [4, 4, 4, 4]
snake_case_ = [3, 3, 3, 3]
if "tiny" in model_name or "small" in model_name or "base" in model_name:
snake_case_ = [3, 3, 3, 3]
if "lrf" in model_name:
snake_case_ = [3, 3, 3, 3]
else:
snake_case_ = [2, 2, 2, 2]
if "tiny" in model_name:
snake_case_ = 96
elif "small" in model_name:
snake_case_ = 96
elif "base" in model_name:
snake_case_ = 128
elif "large" in model_name:
snake_case_ = 192
elif "xlarge" in model_name:
snake_case_ = 256
elif "huge" in model_name:
snake_case_ = 352
# set label information
snake_case_ = '''huggingface/label-files'''
if "large" in model_name or "huge" in model_name:
snake_case_ = '''imagenet-22k-id2label.json'''
else:
snake_case_ = '''imagenet-1k-id2label.json'''
snake_case_ = json.load(open(hf_hub_download(SCREAMING_SNAKE_CASE__ , SCREAMING_SNAKE_CASE__ , repo_type='''dataset''' ) , '''r''' ) )
snake_case_ = {int(SCREAMING_SNAKE_CASE__ ): v for k, v in idalabel.items()}
snake_case_ = {v: k for k, v in idalabel.items()}
snake_case_ = FocalNetConfig(
embed_dim=SCREAMING_SNAKE_CASE__ , depths=SCREAMING_SNAKE_CASE__ , focal_levels=SCREAMING_SNAKE_CASE__ , focal_windows=SCREAMING_SNAKE_CASE__ , use_conv_embed=SCREAMING_SNAKE_CASE__ , idalabel=SCREAMING_SNAKE_CASE__ , labelaid=SCREAMING_SNAKE_CASE__ , use_post_layernorm=SCREAMING_SNAKE_CASE__ , use_layerscale=SCREAMING_SNAKE_CASE__ , )
return config
def __SCREAMING_SNAKE_CASE (SCREAMING_SNAKE_CASE__ ):
if "patch_embed.proj" in name:
snake_case_ = name.replace('''patch_embed.proj''' , '''embeddings.patch_embeddings.projection''' )
if "patch_embed.norm" in name:
snake_case_ = name.replace('''patch_embed.norm''' , '''embeddings.norm''' )
if "layers" in name:
snake_case_ = '''encoder.''' + name
if "encoder.layers" in name:
snake_case_ = name.replace('''encoder.layers''' , '''encoder.stages''' )
if "downsample.proj" in name:
snake_case_ = name.replace('''downsample.proj''' , '''downsample.projection''' )
if "blocks" in name:
snake_case_ = name.replace('''blocks''' , '''layers''' )
if "modulation.f.weight" in name or "modulation.f.bias" in name:
snake_case_ = name.replace('''modulation.f''' , '''modulation.projection_in''' )
if "modulation.h.weight" in name or "modulation.h.bias" in name:
snake_case_ = name.replace('''modulation.h''' , '''modulation.projection_context''' )
if "modulation.proj.weight" in name or "modulation.proj.bias" in name:
snake_case_ = name.replace('''modulation.proj''' , '''modulation.projection_out''' )
if name == "norm.weight":
snake_case_ = '''layernorm.weight'''
if name == "norm.bias":
snake_case_ = '''layernorm.bias'''
if "head" in name:
snake_case_ = name.replace('''head''' , '''classifier''' )
else:
snake_case_ = '''focalnet.''' + name
return name
def __SCREAMING_SNAKE_CASE (SCREAMING_SNAKE_CASE__ , SCREAMING_SNAKE_CASE__ , SCREAMING_SNAKE_CASE__=False ):
# fmt: off
snake_case_ = {
'''focalnet-tiny''': '''https://projects4jw.blob.core.windows.net/focalnet/release/classification/focalnet_tiny_srf.pth''',
'''focalnet-tiny-lrf''': '''https://projects4jw.blob.core.windows.net/focalnet/release/classification/focalnet_tiny_lrf.pth''',
'''focalnet-small''': '''https://projects4jw.blob.core.windows.net/focalnet/release/classification/focalnet_small_srf.pth''',
'''focalnet-small-lrf''': '''https://projects4jw.blob.core.windows.net/focalnet/release/classification/focalnet_small_lrf.pth''',
'''focalnet-base''': '''https://projects4jw.blob.core.windows.net/focalnet/release/classification/focalnet_base_srf.pth''',
'''focalnet-base-lrf''': '''https://projects4jw.blob.core.windows.net/focalnet/release/classification/focalnet_base_lrf.pth''',
'''focalnet-large-lrf-fl3''': '''https://projects4jw.blob.core.windows.net/focalnet/release/classification/focalnet_large_lrf_384.pth''',
'''focalnet-large-lrf-fl4''': '''https://projects4jw.blob.core.windows.net/focalnet/release/classification/focalnet_large_lrf_384_fl4.pth''',
'''focalnet-xlarge-lrf-fl3''': '''https://projects4jw.blob.core.windows.net/focalnet/release/classification/focalnet_xlarge_lrf_384.pth''',
'''focalnet-xlarge-lrf-fl4''': '''https://projects4jw.blob.core.windows.net/focalnet/release/classification/focalnet_xlarge_lrf_384_fl4.pth''',
}
# fmt: on
snake_case_ = model_name_to_url[model_name]
print('''Checkpoint URL: ''' , SCREAMING_SNAKE_CASE__ )
snake_case_ = torch.hub.load_state_dict_from_url(SCREAMING_SNAKE_CASE__ , map_location='''cpu''' )['''model''']
# rename keys
for key in state_dict.copy().keys():
snake_case_ = state_dict.pop(SCREAMING_SNAKE_CASE__ )
snake_case_ = val
snake_case_ = get_focalnet_config(SCREAMING_SNAKE_CASE__ )
snake_case_ = FocalNetForImageClassification(SCREAMING_SNAKE_CASE__ )
model.eval()
# load state dict
model.load_state_dict(SCREAMING_SNAKE_CASE__ )
# verify conversion
snake_case_ = '''http://images.cocodataset.org/val2017/000000039769.jpg'''
snake_case_ = BitImageProcessor(
do_resize=SCREAMING_SNAKE_CASE__ , size={'''shortest_edge''': 256} , resample=PILImageResampling.BILINEAR , do_center_crop=SCREAMING_SNAKE_CASE__ , crop_size=224 , do_normalize=SCREAMING_SNAKE_CASE__ , image_mean=SCREAMING_SNAKE_CASE__ , image_std=SCREAMING_SNAKE_CASE__ , )
snake_case_ = Image.open(requests.get(SCREAMING_SNAKE_CASE__ , stream=SCREAMING_SNAKE_CASE__ ).raw )
snake_case_ = processor(images=SCREAMING_SNAKE_CASE__ , return_tensors='''pt''' )
snake_case_ = transforms.Compose(
[
transforms.Resize(256 ),
transforms.CenterCrop(224 ),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406] , std=[0.229, 0.224, 0.225] ),
] )
snake_case_ = image_transforms(SCREAMING_SNAKE_CASE__ ).unsqueeze(0 )
# verify pixel_values
assert torch.allclose(inputs.pixel_values , SCREAMING_SNAKE_CASE__ , atol=1E-4 )
snake_case_ = model(**SCREAMING_SNAKE_CASE__ )
snake_case_ = outputs.logits.argmax(-1 ).item()
print('''Predicted class:''' , model.config.idalabel[predicted_class_idx] )
print('''First values of logits:''' , outputs.logits[0, :3] )
if model_name == "focalnet-tiny":
snake_case_ = torch.tensor([0.2166, -0.4368, 0.2191] )
elif model_name == "focalnet-tiny-lrf":
snake_case_ = torch.tensor([1.1669, 0.0125, -0.1695] )
elif model_name == "focalnet-small":
snake_case_ = torch.tensor([0.4917, -0.0430, 0.1341] )
elif model_name == "focalnet-small-lrf":
snake_case_ = torch.tensor([-0.2588, -0.5342, -0.2331] )
elif model_name == "focalnet-base":
snake_case_ = torch.tensor([-0.1655, -0.4090, -0.1730] )
elif model_name == "focalnet-base-lrf":
snake_case_ = torch.tensor([0.5306, -0.0483, -0.3928] )
assert torch.allclose(outputs.logits[0, :3] , SCREAMING_SNAKE_CASE__ , atol=1E-4 )
print('''Looks ok!''' )
if pytorch_dump_folder_path is not None:
print(F'''Saving model and processor of {model_name} to {pytorch_dump_folder_path}''' )
model.save_pretrained(SCREAMING_SNAKE_CASE__ )
processor.save_pretrained(SCREAMING_SNAKE_CASE__ )
if push_to_hub:
print(F'''Pushing model and processor of {model_name} to the hub...''' )
model.push_to_hub(F'''{model_name}''' )
processor.push_to_hub(F'''{model_name}''' )
if __name__ == "__main__":
lowerCAmelCase_ = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
'''--model_name''',
default='''focalnet-tiny''',
type=str,
help='''Name of the FocalNet model you\'d like to convert.''',
)
parser.add_argument(
'''--pytorch_dump_folder_path''', default=None, type=str, help='''Path to the output PyTorch model directory.'''
)
parser.add_argument(
'''--push_to_hub''',
action='''store_true''',
help='''Whether to push the model and processor to the hub.''',
)
lowerCAmelCase_ = parser.parse_args()
convert_focalnet_checkpoint(args.model_name, args.pytorch_dump_folder_path, args.push_to_hub) | 39 |
# Lint as: python3
import sys
from collections.abc import Mapping
from typing import TYPE_CHECKING, Dict, Optional
import numpy as np
import pyarrow as pa
from .. import config
from ..utils.logging import get_logger
from ..utils.py_utils import map_nested
from .formatting import TensorFormatter
if TYPE_CHECKING:
import jax
import jaxlib
__a :List[Any] = get_logger()
__a :Optional[dict] = None
class _a ( TensorFormatter[Mapping, 'jax.Array', Mapping] ):
"""simple docstring"""
def __init__( self : str , UpperCAmelCase : int=None , UpperCAmelCase : List[str]=None , **UpperCAmelCase : List[Any] ):
super().__init__(features=UpperCAmelCase )
import jax
from jaxlib.xla_client import Device
if isinstance(UpperCAmelCase , UpperCAmelCase ):
raise ValueError(
f'''Expected {device} to be a `str` not {type(UpperCAmelCase )}, as `jaxlib.xla_extension.Device` '''
"is not serializable neither with `pickle` nor with `dill`. Instead you can surround "
"the device with `str()` to get its string identifier that will be internally mapped "
"to the actual `jaxlib.xla_extension.Device`." )
A_ = device if isinstance(UpperCAmelCase , UpperCAmelCase ) else str(jax.devices()[0] )
# using global variable since `jaxlib.xla_extension.Device` is not serializable neither
# with `pickle` nor with `dill`, so we need to use a global variable instead
global DEVICE_MAPPING
if DEVICE_MAPPING is None:
A_ = self._map_devices_to_str()
if self.device not in list(DEVICE_MAPPING.keys() ):
logger.warning(
f'''Device with string identifier {self.device} not listed among the available '''
f'''devices: {list(DEVICE_MAPPING.keys() )}, so falling back to the default '''
f'''device: {str(jax.devices()[0] )}.''' )
A_ = str(jax.devices()[0] )
A_ = jnp_array_kwargs
@staticmethod
def __A ( ):
import jax
return {str(UpperCAmelCase ): device for device in jax.devices()}
def __A ( self : Optional[int] , UpperCAmelCase : int ):
import jax
import jax.numpy as jnp
if isinstance(UpperCAmelCase , UpperCAmelCase ) and column:
if all(
isinstance(UpperCAmelCase , jax.Array ) and x.shape == column[0].shape and x.dtype == column[0].dtype for x in column ):
return jnp.stack(UpperCAmelCase , axis=0 )
return column
def __A ( self : List[str] , UpperCAmelCase : str ):
import jax
import jax.numpy as jnp
if isinstance(UpperCAmelCase , (str, bytes, type(UpperCAmelCase )) ):
return value
elif isinstance(UpperCAmelCase , (np.character, np.ndarray) ) and np.issubdtype(value.dtype , np.character ):
return value.tolist()
A_ = {}
if isinstance(UpperCAmelCase , (np.number, np.ndarray) ) and np.issubdtype(value.dtype , np.integer ):
# the default int precision depends on the jax config
# see https://jax.readthedocs.io/en/latest/notebooks/Common_Gotchas_in_JAX.html#double-64bit-precision
if jax.config.jax_enable_xaa:
A_ = {"dtype": jnp.intaa}
else:
A_ = {"dtype": jnp.intaa}
elif isinstance(UpperCAmelCase , (np.number, np.ndarray) ) and np.issubdtype(value.dtype , np.floating ):
A_ = {"dtype": jnp.floataa}
elif config.PIL_AVAILABLE and "PIL" in sys.modules:
import PIL.Image
if isinstance(UpperCAmelCase , PIL.Image.Image ):
A_ = np.asarray(UpperCAmelCase )
# using global variable since `jaxlib.xla_extension.Device` is not serializable neither
# with `pickle` nor with `dill`, so we need to use a global variable instead
global DEVICE_MAPPING
if DEVICE_MAPPING is None:
A_ = self._map_devices_to_str()
with jax.default_device(DEVICE_MAPPING[self.device] ):
# calling jnp.array on a np.ndarray does copy the data
# see https://github.com/google/jax/issues/4486
return jnp.array(UpperCAmelCase , **{**default_dtype, **self.jnp_array_kwargs} )
def __A ( self : Any , UpperCAmelCase : Dict ):
import jax
# support for torch, tf, jax etc.
if config.TORCH_AVAILABLE and "torch" in sys.modules:
import torch
if isinstance(UpperCAmelCase , torch.Tensor ):
return self._tensorize(data_struct.detach().cpu().numpy()[()] )
if hasattr(UpperCAmelCase , "__array__" ) and not isinstance(UpperCAmelCase , jax.Array ):
A_ = data_struct.__array__()
# support for nested types like struct of list of struct
if isinstance(UpperCAmelCase , np.ndarray ):
if data_struct.dtype == object: # jax arrays cannot be instantied from an array of objects
return self._consolidate([self.recursive_tensorize(UpperCAmelCase ) for substruct in data_struct] )
elif isinstance(UpperCAmelCase , (list, tuple) ):
return self._consolidate([self.recursive_tensorize(UpperCAmelCase ) for substruct in data_struct] )
return self._tensorize(UpperCAmelCase )
def __A ( self : Tuple , UpperCAmelCase : dict ):
return map_nested(self._recursive_tensorize , UpperCAmelCase , map_list=UpperCAmelCase )
def __A ( self : Dict , UpperCAmelCase : pa.Table ):
A_ = self.numpy_arrow_extractor().extract_row(UpperCAmelCase )
A_ = self.python_features_decoder.decode_row(UpperCAmelCase )
return self.recursive_tensorize(UpperCAmelCase )
def __A ( self : Any , UpperCAmelCase : pa.Table ):
A_ = self.numpy_arrow_extractor().extract_column(UpperCAmelCase )
A_ = self.python_features_decoder.decode_column(UpperCAmelCase , pa_table.column_names[0] )
A_ = self.recursive_tensorize(UpperCAmelCase )
A_ = self._consolidate(UpperCAmelCase )
return column
def __A ( self : Dict , UpperCAmelCase : pa.Table ):
A_ = self.numpy_arrow_extractor().extract_batch(UpperCAmelCase )
A_ = self.python_features_decoder.decode_batch(UpperCAmelCase )
A_ = self.recursive_tensorize(UpperCAmelCase )
for column_name in batch:
A_ = self._consolidate(batch[column_name] )
return batch | 86 | 0 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.